pipecat-satham 0.1.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- pipecat_satham-0.1.0/.gitignore +3 -0
- pipecat_satham-0.1.0/LICENSE +202 -0
- pipecat_satham-0.1.0/PKG-INFO +191 -0
- pipecat_satham-0.1.0/README.md +165 -0
- pipecat_satham-0.1.0/pyproject.toml +38 -0
- pipecat_satham-0.1.0/src/pipecat_satham/__init__.py +16 -0
- pipecat_satham-0.1.0/src/pipecat_satham/py.typed +0 -0
- pipecat_satham-0.1.0/src/pipecat_satham/service.py +556 -0
|
@@ -0,0 +1,202 @@
|
|
|
1
|
+
|
|
2
|
+
Apache License
|
|
3
|
+
Version 2.0, January 2004
|
|
4
|
+
http://www.apache.org/licenses/
|
|
5
|
+
|
|
6
|
+
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
|
|
7
|
+
|
|
8
|
+
1. Definitions.
|
|
9
|
+
|
|
10
|
+
"License" shall mean the terms and conditions for use, reproduction,
|
|
11
|
+
and distribution as defined by Sections 1 through 9 of this document.
|
|
12
|
+
|
|
13
|
+
"Licensor" shall mean the copyright owner or entity authorized by
|
|
14
|
+
the copyright owner that is granting the License.
|
|
15
|
+
|
|
16
|
+
"Legal Entity" shall mean the union of the acting entity and all
|
|
17
|
+
other entities that control, are controlled by, or are under common
|
|
18
|
+
control with that entity. For the purposes of this definition,
|
|
19
|
+
"control" means (i) the power, direct or indirect, to cause the
|
|
20
|
+
direction or management of such entity, whether by contract or
|
|
21
|
+
otherwise, or (ii) ownership of fifty percent (50%) or more of the
|
|
22
|
+
outstanding shares, or (iii) beneficial ownership of such entity.
|
|
23
|
+
|
|
24
|
+
"You" (or "Your") shall mean an individual or Legal Entity
|
|
25
|
+
exercising permissions granted by this License.
|
|
26
|
+
|
|
27
|
+
"Source" form shall mean the preferred form for making modifications,
|
|
28
|
+
including but not limited to software source code, documentation
|
|
29
|
+
source, and configuration files.
|
|
30
|
+
|
|
31
|
+
"Object" form shall mean any form resulting from mechanical
|
|
32
|
+
transformation or translation of a Source form, including but
|
|
33
|
+
not limited to compiled object code, generated documentation,
|
|
34
|
+
and conversions to other media types.
|
|
35
|
+
|
|
36
|
+
"Work" shall mean the work of authorship, whether in Source or
|
|
37
|
+
Object form, made available under the License, as indicated by a
|
|
38
|
+
copyright notice that is included in or attached to the work
|
|
39
|
+
(an example is provided in the Appendix below).
|
|
40
|
+
|
|
41
|
+
"Derivative Works" shall mean any work, whether in Source or Object
|
|
42
|
+
form, that is based on (or derived from) the Work and for which the
|
|
43
|
+
editorial revisions, annotations, elaborations, or other modifications
|
|
44
|
+
represent, as a whole, an original work of authorship. For the purposes
|
|
45
|
+
of this License, Derivative Works shall not include works that remain
|
|
46
|
+
separable from, or merely link (or bind by name) to the interfaces of,
|
|
47
|
+
the Work and Derivative Works thereof.
|
|
48
|
+
|
|
49
|
+
"Contribution" shall mean any work of authorship, including
|
|
50
|
+
the original version of the Work and any modifications or additions
|
|
51
|
+
to that Work or Derivative Works thereof, that is intentionally
|
|
52
|
+
submitted to Licensor for inclusion in the Work by the copyright owner
|
|
53
|
+
or by an individual or Legal Entity authorized to submit on behalf of
|
|
54
|
+
the copyright owner. For the purposes of this definition, "submitted"
|
|
55
|
+
means any form of electronic, verbal, or written communication sent
|
|
56
|
+
to the Licensor or its representatives, including but not limited to
|
|
57
|
+
communication on electronic mailing lists, source code control systems,
|
|
58
|
+
and issue tracking systems that are managed by, or on behalf of, the
|
|
59
|
+
Licensor for the purpose of discussing and improving the Work, but
|
|
60
|
+
excluding communication that is conspicuously marked or otherwise
|
|
61
|
+
designated in writing by the copyright owner as "Not a Contribution."
|
|
62
|
+
|
|
63
|
+
"Contributor" shall mean Licensor and any individual or Legal Entity
|
|
64
|
+
on behalf of whom a Contribution has been received by Licensor and
|
|
65
|
+
subsequently incorporated within the Work.
|
|
66
|
+
|
|
67
|
+
2. Grant of Copyright License. Subject to the terms and conditions of
|
|
68
|
+
this License, each Contributor hereby grants to You a perpetual,
|
|
69
|
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
|
70
|
+
copyright license to reproduce, prepare Derivative Works of,
|
|
71
|
+
publicly display, publicly perform, sublicense, and distribute the
|
|
72
|
+
Work and such Derivative Works in Source or Object form.
|
|
73
|
+
|
|
74
|
+
3. Grant of Patent License. Subject to the terms and conditions of
|
|
75
|
+
this License, each Contributor hereby grants to You a perpetual,
|
|
76
|
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
|
77
|
+
(except as stated in this section) patent license to make, have made,
|
|
78
|
+
use, offer to sell, sell, import, and otherwise transfer the Work,
|
|
79
|
+
where such license applies only to those patent claims licensable
|
|
80
|
+
by such Contributor that are necessarily infringed by their
|
|
81
|
+
Contribution(s) alone or by combination of their Contribution(s)
|
|
82
|
+
with the Work to which such Contribution(s) was submitted. If You
|
|
83
|
+
institute patent litigation against any entity (including a
|
|
84
|
+
cross-claim or counterclaim in a lawsuit) alleging that the Work
|
|
85
|
+
or a Contribution incorporated within the Work constitutes direct
|
|
86
|
+
or contributory patent infringement, then any patent licenses
|
|
87
|
+
granted to You under this License for that Work shall terminate
|
|
88
|
+
as of the date such litigation is filed.
|
|
89
|
+
|
|
90
|
+
4. Redistribution. You may reproduce and distribute copies of the
|
|
91
|
+
Work or Derivative Works thereof in any medium, with or without
|
|
92
|
+
modifications, and in Source or Object form, provided that You
|
|
93
|
+
meet the following conditions:
|
|
94
|
+
|
|
95
|
+
(a) You must give any other recipients of the Work or
|
|
96
|
+
Derivative Works a copy of this License; and
|
|
97
|
+
|
|
98
|
+
(b) You must cause any modified files to carry prominent notices
|
|
99
|
+
stating that You changed the files; and
|
|
100
|
+
|
|
101
|
+
(c) You must retain, in the Source form of any Derivative Works
|
|
102
|
+
that You distribute, all copyright, patent, trademark, and
|
|
103
|
+
attribution notices from the Source form of the Work,
|
|
104
|
+
excluding those notices that do not pertain to any part of
|
|
105
|
+
the Derivative Works; and
|
|
106
|
+
|
|
107
|
+
(d) If the Work includes a "NOTICE" text file as part of its
|
|
108
|
+
distribution, then any Derivative Works that You distribute must
|
|
109
|
+
include a readable copy of the attribution notices contained
|
|
110
|
+
within such NOTICE file, excluding those notices that do not
|
|
111
|
+
pertain to any part of the Derivative Works, in at least one
|
|
112
|
+
of the following places: within a NOTICE text file distributed
|
|
113
|
+
as part of the Derivative Works; within the Source form or
|
|
114
|
+
documentation, if provided along with the Derivative Works; or,
|
|
115
|
+
within a display generated by the Derivative Works, if and
|
|
116
|
+
wherever such third-party notices normally appear. The contents
|
|
117
|
+
of the NOTICE file are for informational purposes only and
|
|
118
|
+
do not modify the License. You may add Your own attribution
|
|
119
|
+
notices within Derivative Works that You distribute, alongside
|
|
120
|
+
or as an addendum to the NOTICE text from the Work, provided
|
|
121
|
+
that such additional attribution notices cannot be construed
|
|
122
|
+
as modifying the License.
|
|
123
|
+
|
|
124
|
+
You may add Your own copyright statement to Your modifications and
|
|
125
|
+
may provide additional or different license terms and conditions
|
|
126
|
+
for use, reproduction, or distribution of Your modifications, or
|
|
127
|
+
for any such Derivative Works as a whole, provided Your use,
|
|
128
|
+
reproduction, and distribution of the Work otherwise complies with
|
|
129
|
+
the conditions stated in this License.
|
|
130
|
+
|
|
131
|
+
5. Submission of Contributions. Unless You explicitly state otherwise,
|
|
132
|
+
any Contribution intentionally submitted for inclusion in the Work
|
|
133
|
+
by You to the Licensor shall be under the terms and conditions of
|
|
134
|
+
this License, without any additional terms or conditions.
|
|
135
|
+
Notwithstanding the above, nothing herein shall supersede or modify
|
|
136
|
+
the terms of any separate license agreement you may have executed
|
|
137
|
+
with Licensor regarding such Contributions.
|
|
138
|
+
|
|
139
|
+
6. Trademarks. This License does not grant permission to use the trade
|
|
140
|
+
names, trademarks, service marks, or product names of the Licensor,
|
|
141
|
+
except as required for reasonable and customary use in describing the
|
|
142
|
+
origin of the Work and reproducing the content of the NOTICE file.
|
|
143
|
+
|
|
144
|
+
7. Disclaimer of Warranty. Unless required by applicable law or
|
|
145
|
+
agreed to in writing, Licensor provides the Work (and each
|
|
146
|
+
Contributor provides its Contributions) on an "AS IS" BASIS,
|
|
147
|
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
|
|
148
|
+
implied, including, without limitation, any warranties or conditions
|
|
149
|
+
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
|
|
150
|
+
PARTICULAR PURPOSE. You are solely responsible for determining the
|
|
151
|
+
appropriateness of using or redistributing the Work and assume any
|
|
152
|
+
risks associated with Your exercise of permissions under this License.
|
|
153
|
+
|
|
154
|
+
8. Limitation of Liability. In no event and under no legal theory,
|
|
155
|
+
whether in tort (including negligence), contract, or otherwise,
|
|
156
|
+
unless required by applicable law (such as deliberate and grossly
|
|
157
|
+
negligent acts) or agreed to in writing, shall any Contributor be
|
|
158
|
+
liable to You for damages, including any direct, indirect, special,
|
|
159
|
+
incidental, or consequential damages of any character arising as a
|
|
160
|
+
result of this License or out of the use or inability to use the
|
|
161
|
+
Work (including but not limited to damages for loss of goodwill,
|
|
162
|
+
work stoppage, computer failure or malfunction, or any and all
|
|
163
|
+
other commercial damages or losses), even if such Contributor
|
|
164
|
+
has been advised of the possibility of such damages.
|
|
165
|
+
|
|
166
|
+
9. Accepting Warranty or Additional Liability. While redistributing
|
|
167
|
+
the Work or Derivative Works thereof, You may choose to offer,
|
|
168
|
+
and charge a fee for, acceptance of support, warranty, indemnity,
|
|
169
|
+
or other liability obligations and/or rights consistent with this
|
|
170
|
+
License. However, in accepting such obligations, You may act only
|
|
171
|
+
on Your own behalf and on Your sole responsibility, not on behalf
|
|
172
|
+
of any other Contributor, and only if You agree to indemnify,
|
|
173
|
+
defend, and hold each Contributor harmless for any liability
|
|
174
|
+
incurred by, or claims asserted against, such Contributor by reason
|
|
175
|
+
of your accepting any such warranty or additional liability.
|
|
176
|
+
|
|
177
|
+
END OF TERMS AND CONDITIONS
|
|
178
|
+
|
|
179
|
+
APPENDIX: How to apply the Apache License to your work.
|
|
180
|
+
|
|
181
|
+
To apply the Apache License to your work, attach the following
|
|
182
|
+
boilerplate notice, with the fields enclosed by brackets "[]"
|
|
183
|
+
replaced with your own identifying information. (Don't include
|
|
184
|
+
the brackets!) The text should be enclosed in the appropriate
|
|
185
|
+
comment syntax for the file format. We also recommend that a
|
|
186
|
+
file or class name and description of purpose be included on the
|
|
187
|
+
same "printed page" as the copyright notice for easier
|
|
188
|
+
identification within third-party archives.
|
|
189
|
+
|
|
190
|
+
Copyright [yyyy] [name of copyright owner]
|
|
191
|
+
|
|
192
|
+
Licensed under the Apache License, Version 2.0 (the "License");
|
|
193
|
+
you may not use this file except in compliance with the License.
|
|
194
|
+
You may obtain a copy of the License at
|
|
195
|
+
|
|
196
|
+
http://www.apache.org/licenses/LICENSE-2.0
|
|
197
|
+
|
|
198
|
+
Unless required by applicable law or agreed to in writing, software
|
|
199
|
+
distributed under the License is distributed on an "AS IS" BASIS,
|
|
200
|
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
201
|
+
See the License for the specific language governing permissions and
|
|
202
|
+
limitations under the License.
|
|
@@ -0,0 +1,191 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: pipecat-satham
|
|
3
|
+
Version: 0.1.0
|
|
4
|
+
Summary: Satham realtime speech-to-speech service for Pipecat
|
|
5
|
+
Project-URL: Homepage, https://satham.ai
|
|
6
|
+
Author: Satham
|
|
7
|
+
License-Expression: Apache-2.0
|
|
8
|
+
License-File: LICENSE
|
|
9
|
+
Keywords: pipecat,pipecat-ai,realtime,satham,speech-to-speech,voice-agent
|
|
10
|
+
Classifier: Development Status :: 3 - Alpha
|
|
11
|
+
Classifier: Framework :: AsyncIO
|
|
12
|
+
Classifier: Intended Audience :: Developers
|
|
13
|
+
Classifier: Operating System :: OS Independent
|
|
14
|
+
Classifier: Programming Language :: Python :: 3
|
|
15
|
+
Classifier: Programming Language :: Python :: 3 :: Only
|
|
16
|
+
Classifier: Programming Language :: Python :: 3.11
|
|
17
|
+
Classifier: Programming Language :: Python :: 3.12
|
|
18
|
+
Classifier: Topic :: Multimedia :: Sound/Audio :: Speech
|
|
19
|
+
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
|
20
|
+
Classifier: Topic :: Software Development :: Libraries :: Python Modules
|
|
21
|
+
Classifier: Typing :: Typed
|
|
22
|
+
Requires-Python: >=3.11
|
|
23
|
+
Requires-Dist: pipecat-ai~=1.16.0
|
|
24
|
+
Requires-Dist: satham~=0.1.0
|
|
25
|
+
Description-Content-Type: text/markdown
|
|
26
|
+
|
|
27
|
+
# pipecat-satham
|
|
28
|
+
|
|
29
|
+
Satham's realtime speech-to-speech model (satham-1) as a Pipecat service. `SathamRealtimeLLMService`
|
|
30
|
+
stands in for STT, LLM and TTS: your context brings the agent's instructions and tools, and Satham
|
|
31
|
+
hears the caller, ends turns, replies in speech and handles barge-in.
|
|
32
|
+
|
|
33
|
+
Pipecat 1.16; Python 3.11+. (Pipecat runs only in Python, so there is no Node version.)
|
|
34
|
+
|
|
35
|
+
```sh
|
|
36
|
+
pip install pipecat-satham
|
|
37
|
+
```
|
|
38
|
+
|
|
39
|
+
## Use
|
|
40
|
+
|
|
41
|
+
You pass your Satham endpoint and API key (or set `SATHAM_URL` and `SATHAM_API_KEY`); there is no
|
|
42
|
+
default server.
|
|
43
|
+
|
|
44
|
+
```python
|
|
45
|
+
import os
|
|
46
|
+
|
|
47
|
+
from pipecat.frames.frames import LLMRunFrame
|
|
48
|
+
from pipecat.pipeline.pipeline import Pipeline
|
|
49
|
+
from pipecat.pipeline.worker import PipelineParams, PipelineWorker
|
|
50
|
+
from pipecat.processors.aggregators.llm_context import LLMContext
|
|
51
|
+
from pipecat.processors.aggregators.llm_response_universal import LLMContextAggregatorPair
|
|
52
|
+
|
|
53
|
+
from pipecat_satham import SathamRealtimeLLMService
|
|
54
|
+
|
|
55
|
+
llm = SathamRealtimeLLMService(
|
|
56
|
+
url=os.environ["SATHAM_URL"],
|
|
57
|
+
api_key=os.environ["SATHAM_API_KEY"],
|
|
58
|
+
greeting="Hello, this is Asha from Example Homes. Is this a good time?",
|
|
59
|
+
language="en",
|
|
60
|
+
language_switch="auto",
|
|
61
|
+
)
|
|
62
|
+
# The context's system message is the call's instructions.
|
|
63
|
+
context = LLMContext(
|
|
64
|
+
[{"role": "system", "content": "You are Asha, who books site visits for Example Homes."}]
|
|
65
|
+
)
|
|
66
|
+
user, assistant = LLMContextAggregatorPair(context)
|
|
67
|
+
pipeline = Pipeline([transport.input(), user, llm, transport.output(), assistant])
|
|
68
|
+
worker = PipelineWorker(pipeline, params=PipelineParams(audio_in_sample_rate=16_000))
|
|
69
|
+
|
|
70
|
+
|
|
71
|
+
@transport.event_handler("on_client_connected")
|
|
72
|
+
async def on_connected(transport, client):
|
|
73
|
+
await worker.queue_frames([LLMRunFrame()]) # starts the call; the agent greets
|
|
74
|
+
```
|
|
75
|
+
|
|
76
|
+
- **The call starts with the first `LLMContextFrame` (or `LLMRunFrame`)**, as with any Pipecat
|
|
77
|
+
realtime service, and the agent speaks first: the `greeting` if there is one, else a reply from its
|
|
78
|
+
instructions. The context's system message is the call's instructions (or pass `instructions=`),
|
|
79
|
+
and the context's tools join the service's. Caller audio before the call starts is dropped.
|
|
80
|
+
- **Audio** goes in as 16 kHz mono; the agent's speech comes out as `TTSAudioRawFrame`s between
|
|
81
|
+
`TTSStartedFrame` and `TTSStoppedFrame`, with its words as `TTSTextFrame`s.
|
|
82
|
+
- **Turns and interruptions.** Satham's speech start and stop become proposed user turns, which the
|
|
83
|
+
context aggregator's external turn strategies turn into user turns and interruptions. An
|
|
84
|
+
interruption from anywhere else in the pipeline cancels the agent's reply.
|
|
85
|
+
- **Transcripts** go upstream as `InterimTranscriptionFrame` / `TranscriptionFrame`; both sides of the
|
|
86
|
+
conversation land in the `LLMContext`.
|
|
87
|
+
- **`modalities=["text"]`** sends the agent's words as `LLMTextFrame`s for a TTS service of your own;
|
|
88
|
+
**`turn_detection=None`** leaves turns to the pipeline's VAD and turn strategies (each
|
|
89
|
+
`UserStoppedSpeakingFrame` ends the caller's turn).
|
|
90
|
+
|
|
91
|
+
## Controls
|
|
92
|
+
|
|
93
|
+
Every keyword of the constructor besides `url`, `api_key`, `instructions`, `tools`, `tool_choice`,
|
|
94
|
+
`connect_timeout` and `start_audio_paused` is a control; other keywords go to Pipecat's `LLMService`.
|
|
95
|
+
Mid-call, `await llm.update_options(...)` changes any control, as does an `LLMUpdateSettingsFrame`
|
|
96
|
+
with `SathamRealtimeLLMService.Settings(controls={...})` (its `temperature`, `top_p` and `max_tokens`
|
|
97
|
+
map to Satham's). Controls you leave out are not sent; values are checked against these ranges, or the
|
|
98
|
+
server's from `GET /v1/info`.
|
|
99
|
+
|
|
100
|
+
The controls marked "needs session controls" need a Satham server with session controls (its
|
|
101
|
+
`GET /v1/info` lists `controls`, as the server does since 2026-11-16); an older server refuses them
|
|
102
|
+
when the call starts. `greeting` and `agent` belong to the start of a call. Setting
|
|
103
|
+
`input_audio_noise_reduction`'s `mode` alone resets `strength_db` to that mode's preset, so send both
|
|
104
|
+
to keep a custom strength; `None` leaves noise reduction at the server's default (on), and `"none"`
|
|
105
|
+
turns it off. `agent` is an optional
|
|
106
|
+
preset on the server and is not needed.
|
|
107
|
+
|
|
108
|
+
<!-- controls -->
|
|
109
|
+
| Option | Range | Default | OpenAI equivalent | Notes |
|
|
110
|
+
| --- | --- | --- | --- | --- |
|
|
111
|
+
| `instructions` | text, up to 16000 characters | none | - | the agent's role; from the LiveKit agent or Pipecat context in the adapters |
|
|
112
|
+
| `greeting` | text, up to 1000 characters | none | - | said first; without one the caller speaks first |
|
|
113
|
+
| `tools` | up to 128 functions | none | - | preview: needs a server with tool support |
|
|
114
|
+
| `tool_choice` | see notes | "auto" | `tool_choice` | "auto", "none", "required" or one function |
|
|
115
|
+
| `agent` | text, up to 64 characters | none | - | optional: a preset agent on the server, such as a demo's |
|
|
116
|
+
| `model` | text, up to 64 characters | "satham-1" | `model` | "satham-1" only today; needs session controls |
|
|
117
|
+
| `modalities` | list | ["audio", "text"] | `modalities` | ["audio", "text"], ["audio"], or ["text"]: words only, for your own TTS; needs session controls |
|
|
118
|
+
| `language` | `auto`, `hi`, `ta`, `te`, `kn`, `bn`, `pa`, `en` | "auto" | - | |
|
|
119
|
+
| `language_switch` | `auto`, `off`, or `None` | "auto" | - | |
|
|
120
|
+
| `voice_on_language_switch` | `keep`, `native`, or `None` | "keep" | - | |
|
|
121
|
+
| `turn_detection.enabled` | `True` / `False` | true | `turn_detection=None` | false (or turn_detection=None): the client ends each turn; needs session controls |
|
|
122
|
+
| `turn_detection.type` | `server_vad`, `semantic_vad` | "server_vad" | `turn_detection.type` | needs session controls |
|
|
123
|
+
| `turn_detection.threshold` | 0.1 to 0.95 | 0.5 | `turn_detection.threshold` | the speech detector's threshold; needs session controls |
|
|
124
|
+
| `turn_detection.eagerness` | `auto`, `low`, `medium`, `high` | "auto" | `turn_detection.eagerness` | semantic_vad's end-of-turn wait; needs session controls |
|
|
125
|
+
| `turn_detection.prefix_padding_ms` | 0 to 1000 | 300 | `turn_detection.prefix_padding_ms` | needs session controls |
|
|
126
|
+
| `turn_detection.min_speech_ms` | 50 to 1000 | 100 | - | needs session controls |
|
|
127
|
+
| `turn_detection.silence_duration_ms` | 150 to 2000 | 350 | `turn_detection.silence_duration_ms` | server_vad's end-of-turn wait; needs session controls |
|
|
128
|
+
| `turn_detection.create_response` | `True` / `False` | true | `turn_detection.create_response` | answer each turn at once; needs session controls |
|
|
129
|
+
| `turn_detection.interrupt_response` | `True` / `False` | true | `turn_detection.interrupt_response` | barge-in on or off; needs session controls |
|
|
130
|
+
| `turn_detection.barge_in.min_speech_ms` | 0 to 2000 | 400 | - | needs session controls |
|
|
131
|
+
| `turn_detection.barge_in.min_probability` | 0 to 1 | 0.8 | - | needs session controls |
|
|
132
|
+
| `turn_detection.barge_in.min_dbfs` | -90 to 0, or `None` | -40 | - | needs session controls |
|
|
133
|
+
| `turn_detection.continue_within_ms` | 0 to 5000, or `None` | 2000 | - | the caller carrying on in the same turn; null: off; needs session controls |
|
|
134
|
+
| `turn_detection.continue_max_heard_ms` | 0 to 5000 | 1000 | - | needs session controls |
|
|
135
|
+
| `input_audio_noise_reduction.enabled` | `True` / `False` | true | `input_audio_noise_reduction={"type": "none"}` | false, "none" or {"type": "none"}: off; needs session controls |
|
|
136
|
+
| `input_audio_noise_reduction.mode` | `near_field`, `far_field` | "near_field" | `input_audio_noise_reduction.type` | resets strength_db to the mode's preset (12 or 30 dB) unless both are sent; needs session controls |
|
|
137
|
+
| `input_audio_noise_reduction.strength_db` | 0 to 100 | 12 | - | needs session controls |
|
|
138
|
+
| `input_audio_noise_reduction.apply_to_transcription` | `True` / `False` | false | - | needs session controls |
|
|
139
|
+
| `input_audio_transcription.enabled` | `True` / `False` | true | `input_audio_transcription=None` | transcript events to the client; needs session controls |
|
|
140
|
+
| `input_audio_transcription.code_mixed` | `True` / `False` | true | - | English words in Latin script, and digits; needs session controls |
|
|
141
|
+
| `voice` | text, up to 64 characters | the language's | `voice` | |
|
|
142
|
+
| `breaths` | `True` / `False`, or `None` | off | - | |
|
|
143
|
+
| `natural_pauses` | `True` / `False`, or `None` | off | - | |
|
|
144
|
+
| `continuation` | `True` / `False`, or `None` | true | - | each phrase carries on from the one before; needs session controls |
|
|
145
|
+
| `speed` | 1 to 1 | 1.0 | `speed` | 1.0 only today; needs session controls |
|
|
146
|
+
| `phrasing.first_phrase_max_chars` | 1 to 120 | 120 | - | at least min_phrase_chars; needs session controls |
|
|
147
|
+
| `phrasing.min_phrase_chars` | 1 to 120 | 20 | - | needs session controls |
|
|
148
|
+
| `normalise_numbers` | `True` / `False` | true | - | numbers and amounts said as words; needs session controls |
|
|
149
|
+
| `pronunciations` | word to respelling | none | - | {word: respelling}, up to 200; needs session controls |
|
|
150
|
+
| `temperature` | 0.6 to 1.2 | 0.7 | `temperature` | needs session controls |
|
|
151
|
+
| `top_p` | 0.05 to 1 | 0.8 | - | needs session controls |
|
|
152
|
+
| `max_response_output_tokens` | 1 to 4096 | 320 | `max_response_output_tokens` | or "inf"; never above the server's; needs session controls |
|
|
153
|
+
| `tone` | `auto`, `conversational`, `happy`, `sad`, `excited`, `professional` | "auto" | - | "auto": the model picks; needs session controls |
|
|
154
|
+
| `repeat_guard` | `True` / `False` | true | - | re-asks a reply that repeats a recent one; needs session controls |
|
|
155
|
+
| `drift_guard` | `True` / `False` | true | - | keeps replies in the call's language; needs session controls |
|
|
156
|
+
<!-- /controls -->
|
|
157
|
+
|
|
158
|
+
## Tools (preview)
|
|
159
|
+
|
|
160
|
+
Preview: tools need a Satham server with tool support.
|
|
161
|
+
|
|
162
|
+
```python
|
|
163
|
+
from pipecat.adapters.llm.schemas.function_schema import FunctionSchema
|
|
164
|
+
from pipecat.adapters.llm.schemas.tools_schema import ToolsSchema
|
|
165
|
+
from pipecat.frames.frames import EndTaskFrame, FunctionCallResultProperties
|
|
166
|
+
from pipecat.processors.frame_processor import FrameDirection
|
|
167
|
+
|
|
168
|
+
END_CALL = FunctionSchema(
|
|
169
|
+
name="end_call",
|
|
170
|
+
description="End the call when the conversation is over. Say a short goodbye in the same turn.",
|
|
171
|
+
properties={"reason": {"type": "string"}},
|
|
172
|
+
required=[],
|
|
173
|
+
)
|
|
174
|
+
|
|
175
|
+
|
|
176
|
+
async def end_call(params):
|
|
177
|
+
await params.result_callback("ended", properties=FunctionCallResultProperties(run_llm=False))
|
|
178
|
+
await params.llm.push_frame(EndTaskFrame(), FrameDirection.UPSTREAM)
|
|
179
|
+
|
|
180
|
+
|
|
181
|
+
llm = SathamRealtimeLLMService(..., tools=ToolsSchema(standard_tools=[END_CALL]))
|
|
182
|
+
llm.register_function("end_call", end_call)
|
|
183
|
+
```
|
|
184
|
+
|
|
185
|
+
Each call the agent makes runs the handler registered with `register_function`; the agent replies to
|
|
186
|
+
its result unless the result carries `FunctionCallResultProperties(run_llm=False)`. A cancelled call
|
|
187
|
+
is answered with `CANCELLED` and no reply. `LLMSetToolsFrame` replaces the tools mid-call.
|
|
188
|
+
|
|
189
|
+
Event handlers: `on_session_updated(service, event)` when the call's language changes, and
|
|
190
|
+
`on_server_event(service, event)` for every Satham event. Not supported: `LLMMessagesAppendFrame`
|
|
191
|
+
(Satham keeps the conversation itself).
|
|
@@ -0,0 +1,165 @@
|
|
|
1
|
+
# pipecat-satham
|
|
2
|
+
|
|
3
|
+
Satham's realtime speech-to-speech model (satham-1) as a Pipecat service. `SathamRealtimeLLMService`
|
|
4
|
+
stands in for STT, LLM and TTS: your context brings the agent's instructions and tools, and Satham
|
|
5
|
+
hears the caller, ends turns, replies in speech and handles barge-in.
|
|
6
|
+
|
|
7
|
+
Pipecat 1.16; Python 3.11+. (Pipecat runs only in Python, so there is no Node version.)
|
|
8
|
+
|
|
9
|
+
```sh
|
|
10
|
+
pip install pipecat-satham
|
|
11
|
+
```
|
|
12
|
+
|
|
13
|
+
## Use
|
|
14
|
+
|
|
15
|
+
You pass your Satham endpoint and API key (or set `SATHAM_URL` and `SATHAM_API_KEY`); there is no
|
|
16
|
+
default server.
|
|
17
|
+
|
|
18
|
+
```python
|
|
19
|
+
import os
|
|
20
|
+
|
|
21
|
+
from pipecat.frames.frames import LLMRunFrame
|
|
22
|
+
from pipecat.pipeline.pipeline import Pipeline
|
|
23
|
+
from pipecat.pipeline.worker import PipelineParams, PipelineWorker
|
|
24
|
+
from pipecat.processors.aggregators.llm_context import LLMContext
|
|
25
|
+
from pipecat.processors.aggregators.llm_response_universal import LLMContextAggregatorPair
|
|
26
|
+
|
|
27
|
+
from pipecat_satham import SathamRealtimeLLMService
|
|
28
|
+
|
|
29
|
+
llm = SathamRealtimeLLMService(
|
|
30
|
+
url=os.environ["SATHAM_URL"],
|
|
31
|
+
api_key=os.environ["SATHAM_API_KEY"],
|
|
32
|
+
greeting="Hello, this is Asha from Example Homes. Is this a good time?",
|
|
33
|
+
language="en",
|
|
34
|
+
language_switch="auto",
|
|
35
|
+
)
|
|
36
|
+
# The context's system message is the call's instructions.
|
|
37
|
+
context = LLMContext(
|
|
38
|
+
[{"role": "system", "content": "You are Asha, who books site visits for Example Homes."}]
|
|
39
|
+
)
|
|
40
|
+
user, assistant = LLMContextAggregatorPair(context)
|
|
41
|
+
pipeline = Pipeline([transport.input(), user, llm, transport.output(), assistant])
|
|
42
|
+
worker = PipelineWorker(pipeline, params=PipelineParams(audio_in_sample_rate=16_000))
|
|
43
|
+
|
|
44
|
+
|
|
45
|
+
@transport.event_handler("on_client_connected")
|
|
46
|
+
async def on_connected(transport, client):
|
|
47
|
+
await worker.queue_frames([LLMRunFrame()]) # starts the call; the agent greets
|
|
48
|
+
```
|
|
49
|
+
|
|
50
|
+
- **The call starts with the first `LLMContextFrame` (or `LLMRunFrame`)**, as with any Pipecat
|
|
51
|
+
realtime service, and the agent speaks first: the `greeting` if there is one, else a reply from its
|
|
52
|
+
instructions. The context's system message is the call's instructions (or pass `instructions=`),
|
|
53
|
+
and the context's tools join the service's. Caller audio before the call starts is dropped.
|
|
54
|
+
- **Audio** goes in as 16 kHz mono; the agent's speech comes out as `TTSAudioRawFrame`s between
|
|
55
|
+
`TTSStartedFrame` and `TTSStoppedFrame`, with its words as `TTSTextFrame`s.
|
|
56
|
+
- **Turns and interruptions.** Satham's speech start and stop become proposed user turns, which the
|
|
57
|
+
context aggregator's external turn strategies turn into user turns and interruptions. An
|
|
58
|
+
interruption from anywhere else in the pipeline cancels the agent's reply.
|
|
59
|
+
- **Transcripts** go upstream as `InterimTranscriptionFrame` / `TranscriptionFrame`; both sides of the
|
|
60
|
+
conversation land in the `LLMContext`.
|
|
61
|
+
- **`modalities=["text"]`** sends the agent's words as `LLMTextFrame`s for a TTS service of your own;
|
|
62
|
+
**`turn_detection=None`** leaves turns to the pipeline's VAD and turn strategies (each
|
|
63
|
+
`UserStoppedSpeakingFrame` ends the caller's turn).
|
|
64
|
+
|
|
65
|
+
## Controls
|
|
66
|
+
|
|
67
|
+
Every keyword of the constructor besides `url`, `api_key`, `instructions`, `tools`, `tool_choice`,
|
|
68
|
+
`connect_timeout` and `start_audio_paused` is a control; other keywords go to Pipecat's `LLMService`.
|
|
69
|
+
Mid-call, `await llm.update_options(...)` changes any control, as does an `LLMUpdateSettingsFrame`
|
|
70
|
+
with `SathamRealtimeLLMService.Settings(controls={...})` (its `temperature`, `top_p` and `max_tokens`
|
|
71
|
+
map to Satham's). Controls you leave out are not sent; values are checked against these ranges, or the
|
|
72
|
+
server's from `GET /v1/info`.
|
|
73
|
+
|
|
74
|
+
The controls marked "needs session controls" need a Satham server with session controls (its
|
|
75
|
+
`GET /v1/info` lists `controls`, as the server does since 2026-11-16); an older server refuses them
|
|
76
|
+
when the call starts. `greeting` and `agent` belong to the start of a call. Setting
|
|
77
|
+
`input_audio_noise_reduction`'s `mode` alone resets `strength_db` to that mode's preset, so send both
|
|
78
|
+
to keep a custom strength; `None` leaves noise reduction at the server's default (on), and `"none"`
|
|
79
|
+
turns it off. `agent` is an optional
|
|
80
|
+
preset on the server and is not needed.
|
|
81
|
+
|
|
82
|
+
<!-- controls -->
|
|
83
|
+
| Option | Range | Default | OpenAI equivalent | Notes |
|
|
84
|
+
| --- | --- | --- | --- | --- |
|
|
85
|
+
| `instructions` | text, up to 16000 characters | none | - | the agent's role; from the LiveKit agent or Pipecat context in the adapters |
|
|
86
|
+
| `greeting` | text, up to 1000 characters | none | - | said first; without one the caller speaks first |
|
|
87
|
+
| `tools` | up to 128 functions | none | - | preview: needs a server with tool support |
|
|
88
|
+
| `tool_choice` | see notes | "auto" | `tool_choice` | "auto", "none", "required" or one function |
|
|
89
|
+
| `agent` | text, up to 64 characters | none | - | optional: a preset agent on the server, such as a demo's |
|
|
90
|
+
| `model` | text, up to 64 characters | "satham-1" | `model` | "satham-1" only today; needs session controls |
|
|
91
|
+
| `modalities` | list | ["audio", "text"] | `modalities` | ["audio", "text"], ["audio"], or ["text"]: words only, for your own TTS; needs session controls |
|
|
92
|
+
| `language` | `auto`, `hi`, `ta`, `te`, `kn`, `bn`, `pa`, `en` | "auto" | - | |
|
|
93
|
+
| `language_switch` | `auto`, `off`, or `None` | "auto" | - | |
|
|
94
|
+
| `voice_on_language_switch` | `keep`, `native`, or `None` | "keep" | - | |
|
|
95
|
+
| `turn_detection.enabled` | `True` / `False` | true | `turn_detection=None` | false (or turn_detection=None): the client ends each turn; needs session controls |
|
|
96
|
+
| `turn_detection.type` | `server_vad`, `semantic_vad` | "server_vad" | `turn_detection.type` | needs session controls |
|
|
97
|
+
| `turn_detection.threshold` | 0.1 to 0.95 | 0.5 | `turn_detection.threshold` | the speech detector's threshold; needs session controls |
|
|
98
|
+
| `turn_detection.eagerness` | `auto`, `low`, `medium`, `high` | "auto" | `turn_detection.eagerness` | semantic_vad's end-of-turn wait; needs session controls |
|
|
99
|
+
| `turn_detection.prefix_padding_ms` | 0 to 1000 | 300 | `turn_detection.prefix_padding_ms` | needs session controls |
|
|
100
|
+
| `turn_detection.min_speech_ms` | 50 to 1000 | 100 | - | needs session controls |
|
|
101
|
+
| `turn_detection.silence_duration_ms` | 150 to 2000 | 350 | `turn_detection.silence_duration_ms` | server_vad's end-of-turn wait; needs session controls |
|
|
102
|
+
| `turn_detection.create_response` | `True` / `False` | true | `turn_detection.create_response` | answer each turn at once; needs session controls |
|
|
103
|
+
| `turn_detection.interrupt_response` | `True` / `False` | true | `turn_detection.interrupt_response` | barge-in on or off; needs session controls |
|
|
104
|
+
| `turn_detection.barge_in.min_speech_ms` | 0 to 2000 | 400 | - | needs session controls |
|
|
105
|
+
| `turn_detection.barge_in.min_probability` | 0 to 1 | 0.8 | - | needs session controls |
|
|
106
|
+
| `turn_detection.barge_in.min_dbfs` | -90 to 0, or `None` | -40 | - | needs session controls |
|
|
107
|
+
| `turn_detection.continue_within_ms` | 0 to 5000, or `None` | 2000 | - | the caller carrying on in the same turn; null: off; needs session controls |
|
|
108
|
+
| `turn_detection.continue_max_heard_ms` | 0 to 5000 | 1000 | - | needs session controls |
|
|
109
|
+
| `input_audio_noise_reduction.enabled` | `True` / `False` | true | `input_audio_noise_reduction={"type": "none"}` | false, "none" or {"type": "none"}: off; needs session controls |
|
|
110
|
+
| `input_audio_noise_reduction.mode` | `near_field`, `far_field` | "near_field" | `input_audio_noise_reduction.type` | resets strength_db to the mode's preset (12 or 30 dB) unless both are sent; needs session controls |
|
|
111
|
+
| `input_audio_noise_reduction.strength_db` | 0 to 100 | 12 | - | needs session controls |
|
|
112
|
+
| `input_audio_noise_reduction.apply_to_transcription` | `True` / `False` | false | - | needs session controls |
|
|
113
|
+
| `input_audio_transcription.enabled` | `True` / `False` | true | `input_audio_transcription=None` | transcript events to the client; needs session controls |
|
|
114
|
+
| `input_audio_transcription.code_mixed` | `True` / `False` | true | - | English words in Latin script, and digits; needs session controls |
|
|
115
|
+
| `voice` | text, up to 64 characters | the language's | `voice` | |
|
|
116
|
+
| `breaths` | `True` / `False`, or `None` | off | - | |
|
|
117
|
+
| `natural_pauses` | `True` / `False`, or `None` | off | - | |
|
|
118
|
+
| `continuation` | `True` / `False`, or `None` | true | - | each phrase carries on from the one before; needs session controls |
|
|
119
|
+
| `speed` | 1 to 1 | 1.0 | `speed` | 1.0 only today; needs session controls |
|
|
120
|
+
| `phrasing.first_phrase_max_chars` | 1 to 120 | 120 | - | at least min_phrase_chars; needs session controls |
|
|
121
|
+
| `phrasing.min_phrase_chars` | 1 to 120 | 20 | - | needs session controls |
|
|
122
|
+
| `normalise_numbers` | `True` / `False` | true | - | numbers and amounts said as words; needs session controls |
|
|
123
|
+
| `pronunciations` | word to respelling | none | - | {word: respelling}, up to 200; needs session controls |
|
|
124
|
+
| `temperature` | 0.6 to 1.2 | 0.7 | `temperature` | needs session controls |
|
|
125
|
+
| `top_p` | 0.05 to 1 | 0.8 | - | needs session controls |
|
|
126
|
+
| `max_response_output_tokens` | 1 to 4096 | 320 | `max_response_output_tokens` | or "inf"; never above the server's; needs session controls |
|
|
127
|
+
| `tone` | `auto`, `conversational`, `happy`, `sad`, `excited`, `professional` | "auto" | - | "auto": the model picks; needs session controls |
|
|
128
|
+
| `repeat_guard` | `True` / `False` | true | - | re-asks a reply that repeats a recent one; needs session controls |
|
|
129
|
+
| `drift_guard` | `True` / `False` | true | - | keeps replies in the call's language; needs session controls |
|
|
130
|
+
<!-- /controls -->
|
|
131
|
+
|
|
132
|
+
## Tools (preview)
|
|
133
|
+
|
|
134
|
+
Preview: tools need a Satham server with tool support.
|
|
135
|
+
|
|
136
|
+
```python
|
|
137
|
+
from pipecat.adapters.llm.schemas.function_schema import FunctionSchema
|
|
138
|
+
from pipecat.adapters.llm.schemas.tools_schema import ToolsSchema
|
|
139
|
+
from pipecat.frames.frames import EndTaskFrame, FunctionCallResultProperties
|
|
140
|
+
from pipecat.processors.frame_processor import FrameDirection
|
|
141
|
+
|
|
142
|
+
END_CALL = FunctionSchema(
|
|
143
|
+
name="end_call",
|
|
144
|
+
description="End the call when the conversation is over. Say a short goodbye in the same turn.",
|
|
145
|
+
properties={"reason": {"type": "string"}},
|
|
146
|
+
required=[],
|
|
147
|
+
)
|
|
148
|
+
|
|
149
|
+
|
|
150
|
+
async def end_call(params):
|
|
151
|
+
await params.result_callback("ended", properties=FunctionCallResultProperties(run_llm=False))
|
|
152
|
+
await params.llm.push_frame(EndTaskFrame(), FrameDirection.UPSTREAM)
|
|
153
|
+
|
|
154
|
+
|
|
155
|
+
llm = SathamRealtimeLLMService(..., tools=ToolsSchema(standard_tools=[END_CALL]))
|
|
156
|
+
llm.register_function("end_call", end_call)
|
|
157
|
+
```
|
|
158
|
+
|
|
159
|
+
Each call the agent makes runs the handler registered with `register_function`; the agent replies to
|
|
160
|
+
its result unless the result carries `FunctionCallResultProperties(run_llm=False)`. A cancelled call
|
|
161
|
+
is answered with `CANCELLED` and no reply. `LLMSetToolsFrame` replaces the tools mid-call.
|
|
162
|
+
|
|
163
|
+
Event handlers: `on_session_updated(service, event)` when the call's language changes, and
|
|
164
|
+
`on_server_event(service, event)` for every Satham event. Not supported: `LLMMessagesAppendFrame`
|
|
165
|
+
(Satham keeps the conversation itself).
|
|
@@ -0,0 +1,38 @@
|
|
|
1
|
+
[project]
|
|
2
|
+
name = "pipecat-satham"
|
|
3
|
+
version = "0.1.0"
|
|
4
|
+
description = "Satham realtime speech-to-speech service for Pipecat"
|
|
5
|
+
readme = "README.md"
|
|
6
|
+
requires-python = ">=3.11"
|
|
7
|
+
license = "Apache-2.0"
|
|
8
|
+
license-files = ["LICENSE"]
|
|
9
|
+
authors = [{ name = "Satham" }]
|
|
10
|
+
keywords = ["satham", "pipecat", "pipecat-ai", "voice-agent", "speech-to-speech", "realtime"]
|
|
11
|
+
classifiers = [
|
|
12
|
+
"Development Status :: 3 - Alpha",
|
|
13
|
+
"Intended Audience :: Developers",
|
|
14
|
+
"Operating System :: OS Independent",
|
|
15
|
+
"Framework :: AsyncIO",
|
|
16
|
+
"Programming Language :: Python :: 3",
|
|
17
|
+
"Programming Language :: Python :: 3 :: Only",
|
|
18
|
+
"Programming Language :: Python :: 3.11",
|
|
19
|
+
"Programming Language :: Python :: 3.12",
|
|
20
|
+
"Topic :: Multimedia :: Sound/Audio :: Speech",
|
|
21
|
+
"Topic :: Scientific/Engineering :: Artificial Intelligence",
|
|
22
|
+
"Topic :: Software Development :: Libraries :: Python Modules",
|
|
23
|
+
"Typing :: Typed",
|
|
24
|
+
]
|
|
25
|
+
dependencies = ["satham~=0.1.0", "pipecat-ai~=1.16.0"]
|
|
26
|
+
|
|
27
|
+
[project.urls]
|
|
28
|
+
Homepage = "https://satham.ai"
|
|
29
|
+
|
|
30
|
+
[build-system]
|
|
31
|
+
requires = ["hatchling==1.27.0"]
|
|
32
|
+
build-backend = "hatchling.build"
|
|
33
|
+
|
|
34
|
+
[tool.hatch.build.targets.wheel]
|
|
35
|
+
packages = ["src/pipecat_satham"]
|
|
36
|
+
|
|
37
|
+
[tool.hatch.build.targets.sdist]
|
|
38
|
+
include = ["src/pipecat_satham", "README.md", "LICENSE"]
|
|
@@ -0,0 +1,16 @@
|
|
|
1
|
+
"""Satham speech-to-speech (satham-1) for Pipecat.
|
|
2
|
+
|
|
3
|
+
from pipecat_satham import SathamRealtimeLLMService
|
|
4
|
+
|
|
5
|
+
llm = SathamRealtimeLLMService(greeting="Hello!") # the role: the context's system message
|
|
6
|
+
"""
|
|
7
|
+
|
|
8
|
+
from .service import INPUT_SAMPLE_RATE, SathamRealtimeLLMService, SathamRealtimeLLMSettings
|
|
9
|
+
|
|
10
|
+
__version__ = "0.1.0"
|
|
11
|
+
__all__ = [
|
|
12
|
+
"INPUT_SAMPLE_RATE",
|
|
13
|
+
"SathamRealtimeLLMService",
|
|
14
|
+
"SathamRealtimeLLMSettings",
|
|
15
|
+
"__version__",
|
|
16
|
+
]
|
|
File without changes
|
|
@@ -0,0 +1,556 @@
|
|
|
1
|
+
"""Satham's speech-to-speech model as a Pipecat service.
|
|
2
|
+
|
|
3
|
+
``SathamRealtimeLLMService`` takes the place of STT, LLM and TTS in a pipeline, in the shape of
|
|
4
|
+
Pipecat's ``OpenAIRealtimeLLMService``: caller audio in (``InputAudioRawFrame``), the agent's
|
|
5
|
+
audio out (``TTSAudioRawFrame``, between ``TTSStartedFrame`` and ``TTSStoppedFrame``), its words
|
|
6
|
+
as ``LLMTextFrame`` and ``TTSTextFrame``, and the caller's words as ``TranscriptionFrame``
|
|
7
|
+
(upstream, for the context aggregator).
|
|
8
|
+
|
|
9
|
+
Satham detects turns and barge-in on the server. Its ``input.speech_started`` and
|
|
10
|
+
``input.speech_stopped`` become ``ProposedUserStarted/StoppedSpeakingFrame``, which the context
|
|
11
|
+
aggregator's external turn strategies turn into user turns and interruptions, so the transport
|
|
12
|
+
stops playing the agent as soon as the caller speaks.
|
|
13
|
+
|
|
14
|
+
A call starts with the first ``LLMContextFrame`` (queue an ``LLMRunFrame`` when the caller
|
|
15
|
+
joins, as with any Pipecat realtime service). The context's system message is the call's
|
|
16
|
+
instructions (unless ``instructions`` is given), and its tools join the service's; the agent then
|
|
17
|
+
speaks first: the ``greeting`` if there is one, else a reply from its instructions
|
|
18
|
+
(``response.create``). Tool calls run the handlers registered with
|
|
19
|
+
``register_function``; their results go back as ``tool.result``, and the server replies unless
|
|
20
|
+
the result asks for no reply (``FunctionCallResultProperties(run_llm=False)``).
|
|
21
|
+
|
|
22
|
+
Every other setting is a control (``satham.Controls``), with OpenAI's names where they exist.
|
|
23
|
+
``modalities=["text"]`` sends the agent's words downstream as ``LLMTextFrame``s for a TTS service
|
|
24
|
+
of your own; ``turn_detection=None`` leaves turns to the pipeline (its VAD and turn strategies),
|
|
25
|
+
and each ``UserStoppedSpeakingFrame`` ends the caller's turn.
|
|
26
|
+
"""
|
|
27
|
+
|
|
28
|
+
from __future__ import annotations
|
|
29
|
+
|
|
30
|
+
import contextlib
|
|
31
|
+
import json
|
|
32
|
+
from dataclasses import dataclass, field
|
|
33
|
+
from typing import Any, Unpack
|
|
34
|
+
|
|
35
|
+
from loguru import logger
|
|
36
|
+
from pipecat.adapters.llm.schemas.tools_schema import ToolsSchema
|
|
37
|
+
from pipecat.audio.utils import create_stream_resampler
|
|
38
|
+
from pipecat.frames.frames import (
|
|
39
|
+
CancelFrame,
|
|
40
|
+
EndFrame,
|
|
41
|
+
Frame,
|
|
42
|
+
FunctionCallCancelFrame,
|
|
43
|
+
FunctionCallFromLLM,
|
|
44
|
+
FunctionCallResultFrame,
|
|
45
|
+
InputAudioRawFrame,
|
|
46
|
+
InterimTranscriptionFrame,
|
|
47
|
+
InterruptionFrame,
|
|
48
|
+
LLMContextFrame,
|
|
49
|
+
LLMFullResponseEndFrame,
|
|
50
|
+
LLMFullResponseStartFrame,
|
|
51
|
+
LLMMessagesAppendFrame,
|
|
52
|
+
LLMRunFrame,
|
|
53
|
+
LLMServiceMetadataFrame,
|
|
54
|
+
LLMSetToolsFrame,
|
|
55
|
+
LLMTextFrame,
|
|
56
|
+
ProposedUserStartedSpeakingFrame,
|
|
57
|
+
ProposedUserStoppedSpeakingFrame,
|
|
58
|
+
StartFrame,
|
|
59
|
+
TextType,
|
|
60
|
+
TranscriptionFrame,
|
|
61
|
+
TTSAudioRawFrame,
|
|
62
|
+
TTSStartedFrame,
|
|
63
|
+
TTSStoppedFrame,
|
|
64
|
+
TTSTextFrame,
|
|
65
|
+
UserStoppedSpeakingFrame,
|
|
66
|
+
)
|
|
67
|
+
from pipecat.processors.aggregators.llm_context import LLMContext
|
|
68
|
+
from pipecat.processors.frame_processor import FrameDirection
|
|
69
|
+
from pipecat.services.llm_service import LLMService
|
|
70
|
+
from pipecat.services.settings import LLMSettings
|
|
71
|
+
from pipecat.transcriptions.language import Language
|
|
72
|
+
from pipecat.turns.user_turn_strategies import ExternalUserTurnStrategies
|
|
73
|
+
from pipecat.utils.time import time_now_iso8601
|
|
74
|
+
from pipecat.utils.types import NOT_GIVEN, NotGiven, is_given
|
|
75
|
+
from satham.controls import to_wire
|
|
76
|
+
|
|
77
|
+
import satham
|
|
78
|
+
from satham import _protocol as wire
|
|
79
|
+
|
|
80
|
+
INPUT_SAMPLE_RATE = 16_000
|
|
81
|
+
"""Caller audio goes to Satham as 16 kHz mono 16-bit PCM."""
|
|
82
|
+
|
|
83
|
+
_FROM_LLM_SETTINGS = {
|
|
84
|
+
"temperature": "temperature",
|
|
85
|
+
"top_p": "top_p",
|
|
86
|
+
"max_tokens": "max_response_output_tokens",
|
|
87
|
+
}
|
|
88
|
+
|
|
89
|
+
|
|
90
|
+
@dataclass
|
|
91
|
+
class SathamRealtimeLLMSettings(LLMSettings):
|
|
92
|
+
"""The settings a running call can change (``LLMUpdateSettingsFrame``, ``session.update``).
|
|
93
|
+
|
|
94
|
+
Parameters:
|
|
95
|
+
controls: any of Satham's controls (``satham.Controls``), e.g. ``{"speed": 1.1,
|
|
96
|
+
"turn_detection": {"silence_duration_ms": 500}}``; only the given ones change.
|
|
97
|
+
|
|
98
|
+
The common ``temperature``, ``top_p`` and ``max_tokens`` map to Satham's ``temperature``,
|
|
99
|
+
``top_p`` and ``max_response_output_tokens``.
|
|
100
|
+
"""
|
|
101
|
+
|
|
102
|
+
controls: dict[str, Any] | NotGiven = field(default_factory=lambda: NOT_GIVEN)
|
|
103
|
+
|
|
104
|
+
|
|
105
|
+
def _tool_dicts(tools: ToolsSchema | list[Any] | NotGiven | None) -> list[dict[str, Any]]:
|
|
106
|
+
"""Pipecat tools as OpenAI function definitions."""
|
|
107
|
+
if not is_given(tools) or tools is None:
|
|
108
|
+
return []
|
|
109
|
+
if isinstance(tools, ToolsSchema):
|
|
110
|
+
return [{"type": "function", "function": t.to_default_dict()} for t in tools.standard_tools]
|
|
111
|
+
out: list[dict[str, Any]] = []
|
|
112
|
+
for tool in tools:
|
|
113
|
+
if hasattr(tool, "to_default_dict"):
|
|
114
|
+
out.append({"type": "function", "function": tool.to_default_dict()})
|
|
115
|
+
elif isinstance(tool, dict):
|
|
116
|
+
out.append(tool)
|
|
117
|
+
return out
|
|
118
|
+
|
|
119
|
+
|
|
120
|
+
def _language(code: str | None) -> Language | None:
|
|
121
|
+
try:
|
|
122
|
+
return Language(code) if code else None
|
|
123
|
+
except ValueError:
|
|
124
|
+
return None
|
|
125
|
+
|
|
126
|
+
|
|
127
|
+
class SathamRealtimeLLMService(LLMService):
|
|
128
|
+
"""Satham (satham-1) speech-to-speech for Pipecat pipelines.
|
|
129
|
+
|
|
130
|
+
Args:
|
|
131
|
+
url: your Satham endpoint (``wss://...`` or ``https://...``); ``$SATHAM_URL``.
|
|
132
|
+
api_key: your API key; ``$SATHAM_API_KEY``.
|
|
133
|
+
instructions: the agent's role; by default, the context's system message.
|
|
134
|
+
greeting: the opening line; without one the agent opens with a reply of its own.
|
|
135
|
+
language: ``"auto"`` or one of ``hi ta te kn bn pa en``.
|
|
136
|
+
voice: a named voice, ``"default"``, or a cloned voice's id.
|
|
137
|
+
breaths: keep the voice's breaths.
|
|
138
|
+
natural_pauses: keep the voice's longer pauses inside a phrase.
|
|
139
|
+
language_switch: ``"auto"`` or ``"off"``.
|
|
140
|
+
voice_on_language_switch: ``"keep"`` or ``"native"``.
|
|
141
|
+
tools: functions the agent may call, beside the context's (``ToolsSchema`` or a list).
|
|
142
|
+
tool_choice: ``"auto"``, ``"none"``, ``"required"``, or one function as
|
|
143
|
+
``{"type": "function", "name": ...}``.
|
|
144
|
+
connect_timeout: seconds to wait for the call to start.
|
|
145
|
+
start_audio_paused: drop caller audio until ``set_audio_input_paused(False)``.
|
|
146
|
+
agent: optional, a preset agent on the server (such as a demo's).
|
|
147
|
+
|
|
148
|
+
Event handlers: ``on_session_updated(service, event)`` when the call's language changes
|
|
149
|
+
(``event`` is a ``satham.SessionUpdated``), ``on_server_event(service, event)`` for every
|
|
150
|
+
Satham event.
|
|
151
|
+
"""
|
|
152
|
+
|
|
153
|
+
Settings = SathamRealtimeLLMSettings
|
|
154
|
+
_settings: SathamRealtimeLLMSettings
|
|
155
|
+
|
|
156
|
+
def __init__(
|
|
157
|
+
self,
|
|
158
|
+
*,
|
|
159
|
+
url: str | None = None,
|
|
160
|
+
api_key: str | None = None,
|
|
161
|
+
instructions: str | None = None,
|
|
162
|
+
tools: ToolsSchema | list[Any] | None = None,
|
|
163
|
+
tool_choice: str | dict[str, Any] | None = None,
|
|
164
|
+
connect_timeout: float = 10.0,
|
|
165
|
+
start_audio_paused: bool = False,
|
|
166
|
+
settings: SathamRealtimeLLMSettings | None = None,
|
|
167
|
+
**controls: Unpack[satham.Controls],
|
|
168
|
+
) -> None:
|
|
169
|
+
defaults = SathamRealtimeLLMSettings(
|
|
170
|
+
model=str(controls.get("model") or "satham-1"),
|
|
171
|
+
system_instruction=instructions,
|
|
172
|
+
temperature=None,
|
|
173
|
+
max_tokens=None,
|
|
174
|
+
top_p=None,
|
|
175
|
+
top_k=None,
|
|
176
|
+
frequency_penalty=None,
|
|
177
|
+
presence_penalty=None,
|
|
178
|
+
seed=None,
|
|
179
|
+
filter_incomplete_user_turns=False,
|
|
180
|
+
user_turn_completion_config=None,
|
|
181
|
+
controls={},
|
|
182
|
+
)
|
|
183
|
+
if settings is not None:
|
|
184
|
+
defaults.apply_update(settings)
|
|
185
|
+
# Keywords that are not Satham controls go to Pipecat's LLMService (run_in_parallel, ...).
|
|
186
|
+
known = satham.Controls.__optional_keys__ | satham.Controls.__required_keys__
|
|
187
|
+
service_kwargs = {k: v for k, v in controls.items() if k not in known}
|
|
188
|
+
super().__init__(settings=defaults, **service_kwargs)
|
|
189
|
+
self._client = satham.Client(url, api_key, open_timeout=connect_timeout)
|
|
190
|
+
self._controls: dict[str, Any] = {k: v for k, v in controls.items() if k in known}
|
|
191
|
+
# Check the controls now, not when the first call starts.
|
|
192
|
+
to_wire(self._controls, where="session.start", limits=self._client.limits)
|
|
193
|
+
self._tools = tools
|
|
194
|
+
self._tool_choice = tool_choice
|
|
195
|
+
self._text_only = list(self._controls.get("modalities") or ["text", "audio"]) == ["text"]
|
|
196
|
+
turns = self._controls.get("turn_detection", {})
|
|
197
|
+
turns = turns.model_dump(exclude_none=True) if hasattr(turns, "model_dump") else turns
|
|
198
|
+
self._manual_turns = "turn_detection" in self._controls and (
|
|
199
|
+
turns is None or (isinstance(turns, dict) and turns.get("enabled") is False)
|
|
200
|
+
)
|
|
201
|
+
self._audio_paused = start_audio_paused
|
|
202
|
+
self._resampler = create_stream_resampler()
|
|
203
|
+
self._context: LLMContext | None = None
|
|
204
|
+
self._call: satham.Session | None = None
|
|
205
|
+
self._starting = False
|
|
206
|
+
self._receive_task: Any = None
|
|
207
|
+
self._response_id: str | None = None
|
|
208
|
+
self._audio_started = False
|
|
209
|
+
self._interrupted_by_server = False
|
|
210
|
+
self._barged_in = False
|
|
211
|
+
self._proposed = 0
|
|
212
|
+
self._spoken_over: str | None = None
|
|
213
|
+
self._open_calls: set[str] = set()
|
|
214
|
+
language = self._controls.get("language")
|
|
215
|
+
self.language: str | None = None if language in (None, "auto") else str(language)
|
|
216
|
+
self._register_event_handler("on_session_updated")
|
|
217
|
+
self._register_event_handler("on_server_event")
|
|
218
|
+
|
|
219
|
+
# -- Pipecat hooks -----------------------------------------------------------------------
|
|
220
|
+
|
|
221
|
+
def can_generate_metrics(self) -> bool:
|
|
222
|
+
return True
|
|
223
|
+
|
|
224
|
+
def service_metadata_frame(self) -> LLMServiceMetadataFrame:
|
|
225
|
+
return LLMServiceMetadataFrame(
|
|
226
|
+
service_name=self.name,
|
|
227
|
+
is_realtime_service=True,
|
|
228
|
+
# With manual turns the pipeline's own VAD and turn strategies decide.
|
|
229
|
+
user_turn_strategies=None if self._manual_turns else ExternalUserTurnStrategies(),
|
|
230
|
+
)
|
|
231
|
+
|
|
232
|
+
def _service_tools(self) -> ToolsSchema | list[Any] | None:
|
|
233
|
+
return self._tools
|
|
234
|
+
|
|
235
|
+
def set_audio_input_paused(self, paused: bool) -> None:
|
|
236
|
+
self._audio_paused = paused
|
|
237
|
+
|
|
238
|
+
@property
|
|
239
|
+
def call(self) -> satham.Session | None:
|
|
240
|
+
"""The underlying ``satham.Session``, once the call has started."""
|
|
241
|
+
return self._call
|
|
242
|
+
|
|
243
|
+
async def start(self, frame: StartFrame) -> None:
|
|
244
|
+
await super().start(frame)
|
|
245
|
+
|
|
246
|
+
async def stop(self, frame: EndFrame) -> None:
|
|
247
|
+
await super().stop(frame)
|
|
248
|
+
await self._end_call()
|
|
249
|
+
|
|
250
|
+
async def cancel(self, frame: CancelFrame) -> None:
|
|
251
|
+
await super().cancel(frame)
|
|
252
|
+
await self._end_call()
|
|
253
|
+
|
|
254
|
+
async def cleanup(self) -> None:
|
|
255
|
+
await super().cleanup()
|
|
256
|
+
await self._end_call()
|
|
257
|
+
|
|
258
|
+
async def process_frame(self, frame: Frame, direction: FrameDirection) -> None:
|
|
259
|
+
await super().process_frame(frame, direction)
|
|
260
|
+
if isinstance(frame, LLMContextFrame):
|
|
261
|
+
self._context = frame.context
|
|
262
|
+
await self._start_call()
|
|
263
|
+
elif isinstance(frame, LLMRunFrame):
|
|
264
|
+
await self._start_call()
|
|
265
|
+
elif isinstance(frame, InputAudioRawFrame):
|
|
266
|
+
if not self._audio_paused:
|
|
267
|
+
await self._send_user_audio(frame)
|
|
268
|
+
elif isinstance(frame, InterruptionFrame):
|
|
269
|
+
await self._handle_interruption()
|
|
270
|
+
elif isinstance(frame, UserStoppedSpeakingFrame) and self._manual_turns:
|
|
271
|
+
await self._send({"type": wire.INPUT_COMMIT}) # the pipeline ended the caller's turn
|
|
272
|
+
elif isinstance(frame, LLMSetToolsFrame):
|
|
273
|
+
self._sync_registered_tool_handlers(frame.tools)
|
|
274
|
+
self._tools = frame.tools
|
|
275
|
+
await self._send({"type": wire.SESSION_UPDATE, "tools": self._all_tools()})
|
|
276
|
+
elif isinstance(frame, LLMMessagesAppendFrame):
|
|
277
|
+
logger.warning(
|
|
278
|
+
f"{self}: Satham keeps the conversation itself; appended messages are ignored"
|
|
279
|
+
)
|
|
280
|
+
await self.push_frame(frame, direction)
|
|
281
|
+
|
|
282
|
+
async def push_frame(
|
|
283
|
+
self, frame: Frame, direction: FrameDirection = FrameDirection.DOWNSTREAM
|
|
284
|
+
) -> None:
|
|
285
|
+
if direction == FrameDirection.DOWNSTREAM:
|
|
286
|
+
if isinstance(frame, FunctionCallResultFrame):
|
|
287
|
+
await self._send_tool_result(frame)
|
|
288
|
+
elif (
|
|
289
|
+
isinstance(frame, FunctionCallCancelFrame)
|
|
290
|
+
and frame.tool_call_id in self._open_calls
|
|
291
|
+
):
|
|
292
|
+
self._open_calls.discard(frame.tool_call_id)
|
|
293
|
+
await self._send(wire.tool_result(frame.tool_call_id, "CANCELLED", respond=False))
|
|
294
|
+
await super().push_frame(frame, direction)
|
|
295
|
+
|
|
296
|
+
async def _update_settings(self, delta: LLMSettings) -> dict[str, Any]:
|
|
297
|
+
changed = await super()._update_settings(delta)
|
|
298
|
+
changes: dict[str, Any] = {}
|
|
299
|
+
if "controls" in changed and is_given(getattr(delta, "controls", NOT_GIVEN)):
|
|
300
|
+
changes.update(delta.controls)
|
|
301
|
+
for name, control in _FROM_LLM_SETTINGS.items():
|
|
302
|
+
if name in changed:
|
|
303
|
+
changes[control] = getattr(self._settings, name)
|
|
304
|
+
if changes:
|
|
305
|
+
await self.update_options(**changes)
|
|
306
|
+
return changed
|
|
307
|
+
|
|
308
|
+
async def update_options(self, **changes: Unpack[satham.Controls]) -> None:
|
|
309
|
+
"""Change the call's controls (``session.update``); before it starts, its start."""
|
|
310
|
+
fields = to_wire(dict(changes), where="session.update", limits=self._client.limits)
|
|
311
|
+
if self._call is None:
|
|
312
|
+
self._controls.update(changes)
|
|
313
|
+
return
|
|
314
|
+
message = wire.session_update(fields)
|
|
315
|
+
if message is not None:
|
|
316
|
+
await self._send(message)
|
|
317
|
+
if changes.get("language") not in (None, "auto"):
|
|
318
|
+
self.language = str(changes["language"])
|
|
319
|
+
|
|
320
|
+
# -- the call ----------------------------------------------------------------------------
|
|
321
|
+
|
|
322
|
+
def _all_tools(self) -> list[dict[str, Any]]:
|
|
323
|
+
tools = {
|
|
324
|
+
t["function"]["name"] if "function" in t else t["name"]: t
|
|
325
|
+
for t in _tool_dicts(self._tools)
|
|
326
|
+
}
|
|
327
|
+
if self._context is not None:
|
|
328
|
+
for tool in _tool_dicts(self._context.tools):
|
|
329
|
+
name = tool["function"]["name"] if "function" in tool else tool["name"]
|
|
330
|
+
tools.setdefault(name, tool)
|
|
331
|
+
return list(tools.values())
|
|
332
|
+
|
|
333
|
+
def _context_instructions(self) -> str | None:
|
|
334
|
+
if self._context is None:
|
|
335
|
+
return None
|
|
336
|
+
for message in self._context.get_messages():
|
|
337
|
+
if isinstance(message, dict) and message.get("role") in ("system", "developer"):
|
|
338
|
+
content = message.get("content")
|
|
339
|
+
if isinstance(content, str) and content.strip():
|
|
340
|
+
return content
|
|
341
|
+
if isinstance(content, list):
|
|
342
|
+
text = " ".join(p.get("text", "") for p in content if isinstance(p, dict))
|
|
343
|
+
if text.strip():
|
|
344
|
+
return text
|
|
345
|
+
return None
|
|
346
|
+
|
|
347
|
+
async def _start_call(self) -> None:
|
|
348
|
+
if self._call is not None or self._starting:
|
|
349
|
+
return
|
|
350
|
+
self._starting = True
|
|
351
|
+
system = self._settings.system_instruction
|
|
352
|
+
instructions = (system if is_given(system) else None) or self._context_instructions()
|
|
353
|
+
tools = self._all_tools()
|
|
354
|
+
try:
|
|
355
|
+
with contextlib.suppress(satham.SathamError):
|
|
356
|
+
await self._client.load_controls() # the server's ranges, when it lists them
|
|
357
|
+
self._call = await self._client.connect(
|
|
358
|
+
**self._controls,
|
|
359
|
+
instructions=instructions,
|
|
360
|
+
sample_rate=INPUT_SAMPLE_RATE,
|
|
361
|
+
tools=tools or None,
|
|
362
|
+
tool_choice=self._tool_choice if tools else None,
|
|
363
|
+
)
|
|
364
|
+
except (satham.SathamError, ValueError) as e:
|
|
365
|
+
await self.push_error(error_msg=f"Satham: {e}", exception=e, fatal=True)
|
|
366
|
+
return
|
|
367
|
+
finally:
|
|
368
|
+
self._starting = False
|
|
369
|
+
if self._call.language:
|
|
370
|
+
self.language = self._call.language
|
|
371
|
+
self._receive_task = self.create_task(self._receive(self._call), name="satham-receive")
|
|
372
|
+
if "greeting" not in self._controls and "agent" not in self._controls:
|
|
373
|
+
# As with Pipecat's OpenAI realtime service, the run frame has the agent speak first.
|
|
374
|
+
await self._send(wire.response_create())
|
|
375
|
+
|
|
376
|
+
async def _end_call(self) -> None:
|
|
377
|
+
call, self._call = self._call, None
|
|
378
|
+
if call is not None:
|
|
379
|
+
await call.close()
|
|
380
|
+
if self._receive_task is not None:
|
|
381
|
+
await self.cancel_task(self._receive_task, timeout=1.0)
|
|
382
|
+
self._receive_task = None
|
|
383
|
+
|
|
384
|
+
async def _send(self, message: dict[str, Any]) -> None:
|
|
385
|
+
if self._call is None:
|
|
386
|
+
return
|
|
387
|
+
with contextlib.suppress(satham.SessionClosedError):
|
|
388
|
+
await self._call.send(message)
|
|
389
|
+
|
|
390
|
+
async def _send_user_audio(self, frame: InputAudioRawFrame) -> None:
|
|
391
|
+
if self._call is None:
|
|
392
|
+
return # no call yet: the caller has not been put through
|
|
393
|
+
audio = frame.audio
|
|
394
|
+
if frame.num_channels == 2:
|
|
395
|
+
samples = memoryview(audio).cast("h")
|
|
396
|
+
audio = b"".join(
|
|
397
|
+
((samples[i] + samples[i + 1]) // 2).to_bytes(2, "little", signed=True)
|
|
398
|
+
for i in range(0, len(samples) - 1, 2)
|
|
399
|
+
)
|
|
400
|
+
if frame.sample_rate != INPUT_SAMPLE_RATE:
|
|
401
|
+
audio = await self._resampler.resample(audio, frame.sample_rate, INPUT_SAMPLE_RATE)
|
|
402
|
+
if audio:
|
|
403
|
+
with contextlib.suppress(satham.SessionClosedError):
|
|
404
|
+
await self._call.send_audio(audio)
|
|
405
|
+
|
|
406
|
+
async def _send_tool_result(self, frame: FunctionCallResultFrame) -> None:
|
|
407
|
+
if frame.tool_call_id not in self._open_calls:
|
|
408
|
+
return
|
|
409
|
+
if frame.properties is not None and not frame.properties.is_final:
|
|
410
|
+
return # Satham takes one result per call: the final one
|
|
411
|
+
self._open_calls.discard(frame.tool_call_id)
|
|
412
|
+
run_llm = frame.run_llm
|
|
413
|
+
if run_llm is None and frame.properties is not None:
|
|
414
|
+
run_llm = frame.properties.run_llm
|
|
415
|
+
result = frame.result
|
|
416
|
+
if frame.error:
|
|
417
|
+
output = f"error: {frame.error}"
|
|
418
|
+
elif result is None or isinstance(result, str):
|
|
419
|
+
output = result or ""
|
|
420
|
+
else:
|
|
421
|
+
output = json.dumps(result, ensure_ascii=False, default=str)
|
|
422
|
+
await self._send(wire.tool_result(frame.tool_call_id, output, respond=run_llm is not False))
|
|
423
|
+
|
|
424
|
+
async def _handle_interruption(self) -> None:
|
|
425
|
+
if self._proposed:
|
|
426
|
+
# The interruption that the caller's speech (proposed by this service) brought
|
|
427
|
+
# about: the server has stopped the reply already. A reply started since then is
|
|
428
|
+
# the answer to that speech, and plays on.
|
|
429
|
+
self._proposed -= 1
|
|
430
|
+
if self._response_id is not None and self._response_id == self._spoken_over:
|
|
431
|
+
await self._end_response()
|
|
432
|
+
return
|
|
433
|
+
if self._response_id is None:
|
|
434
|
+
return
|
|
435
|
+
if not self._interrupted_by_server:
|
|
436
|
+
await self._send({"type": wire.RESPONSE_CANCEL})
|
|
437
|
+
await self._end_response()
|
|
438
|
+
|
|
439
|
+
async def _propose_user_turn(self) -> None:
|
|
440
|
+
self._proposed += 1
|
|
441
|
+
self._spoken_over = self._response_id
|
|
442
|
+
await self.broadcast_frame(ProposedUserStartedSpeakingFrame)
|
|
443
|
+
|
|
444
|
+
async def _end_response(self) -> None:
|
|
445
|
+
if self._response_id is None:
|
|
446
|
+
return
|
|
447
|
+
self._response_id = None
|
|
448
|
+
await self.stop_all_metrics()
|
|
449
|
+
if self._audio_started:
|
|
450
|
+
self._audio_started = False
|
|
451
|
+
await self.push_frame(TTSStoppedFrame())
|
|
452
|
+
await self.push_frame(LLMFullResponseEndFrame())
|
|
453
|
+
|
|
454
|
+
# -- server events -----------------------------------------------------------------------
|
|
455
|
+
|
|
456
|
+
async def _receive(self, call: satham.Session) -> None:
|
|
457
|
+
async for event in call:
|
|
458
|
+
await self._call_event_handler("on_server_event", event)
|
|
459
|
+
await self._on_event(event)
|
|
460
|
+
if self._call is call:
|
|
461
|
+
reason = call.end_reason or f"connection closed ({call.close_code})"
|
|
462
|
+
logger.info(f"{self}: the Satham call ended: {reason}")
|
|
463
|
+
|
|
464
|
+
async def _on_event(self, event: satham.Event) -> None:
|
|
465
|
+
if isinstance(event, satham.ResponseAudio):
|
|
466
|
+
if self._text_only or event.response_id != self._response_id or not event.data:
|
|
467
|
+
return
|
|
468
|
+
if not self._audio_started:
|
|
469
|
+
self._audio_started = True
|
|
470
|
+
await self.stop_ttfb_metrics()
|
|
471
|
+
await self.push_frame(TTSStartedFrame())
|
|
472
|
+
fmt = event.format
|
|
473
|
+
await self.push_frame(
|
|
474
|
+
TTSAudioRawFrame(
|
|
475
|
+
audio=event.data, sample_rate=fmt.sample_rate, num_channels=fmt.channels
|
|
476
|
+
)
|
|
477
|
+
)
|
|
478
|
+
elif isinstance(event, satham.ResponseTextDelta):
|
|
479
|
+
if event.response_id == self._response_id and event.text and self._text_only:
|
|
480
|
+
await self.push_frame(LLMTextFrame(event.text)) # for the pipeline's own TTS
|
|
481
|
+
elif event.response_id == self._response_id and event.text:
|
|
482
|
+
text = LLMTextFrame(event.text)
|
|
483
|
+
text.append_to_context = False
|
|
484
|
+
await self.push_frame(text)
|
|
485
|
+
spoken = TTSTextFrame(event.text, text_type=TextType.SENTENCE)
|
|
486
|
+
spoken.includes_inter_frame_spaces = True
|
|
487
|
+
await self.push_frame(spoken)
|
|
488
|
+
elif isinstance(event, satham.SpeechStarted):
|
|
489
|
+
if self._barged_in:
|
|
490
|
+
self._barged_in = False # already proposed when the reply was cut off
|
|
491
|
+
else:
|
|
492
|
+
await self._propose_user_turn()
|
|
493
|
+
elif isinstance(event, satham.SpeechStopped):
|
|
494
|
+
await self.start_processing_metrics()
|
|
495
|
+
await self.start_ttfb_metrics()
|
|
496
|
+
await self.broadcast_frame(ProposedUserStoppedSpeakingFrame)
|
|
497
|
+
elif isinstance(event, satham.TranscriptPartial):
|
|
498
|
+
await self.push_frame(
|
|
499
|
+
InterimTranscriptionFrame(
|
|
500
|
+
event.text, "", time_now_iso8601(), _language(event.language), result=event.raw
|
|
501
|
+
),
|
|
502
|
+
FrameDirection.UPSTREAM,
|
|
503
|
+
)
|
|
504
|
+
elif isinstance(event, satham.TranscriptFinal):
|
|
505
|
+
if event.text:
|
|
506
|
+
await self.push_frame(
|
|
507
|
+
TranscriptionFrame(
|
|
508
|
+
event.text,
|
|
509
|
+
"",
|
|
510
|
+
time_now_iso8601(),
|
|
511
|
+
_language(event.language),
|
|
512
|
+
result=event.raw,
|
|
513
|
+
),
|
|
514
|
+
FrameDirection.UPSTREAM,
|
|
515
|
+
)
|
|
516
|
+
elif isinstance(event, satham.ResponseStarted):
|
|
517
|
+
await self._end_response()
|
|
518
|
+
self._response_id = event.response_id
|
|
519
|
+
self._interrupted_by_server = False
|
|
520
|
+
await self.push_frame(LLMFullResponseStartFrame())
|
|
521
|
+
elif isinstance(event, satham.ResponseDone):
|
|
522
|
+
if event.response_id == self._response_id:
|
|
523
|
+
await self._end_response()
|
|
524
|
+
elif isinstance(event, satham.ResponseInterrupted):
|
|
525
|
+
if event.response_id == self._response_id:
|
|
526
|
+
self._interrupted_by_server = True
|
|
527
|
+
if event.reason == "barge_in":
|
|
528
|
+
# The caller is speaking; propose the turn now, while the reply is still
|
|
529
|
+
# open, so the interruption reaches the transport before more audio does.
|
|
530
|
+
self._barged_in = True
|
|
531
|
+
await self._propose_user_turn()
|
|
532
|
+
await self._end_response()
|
|
533
|
+
elif isinstance(event, satham.ToolCall):
|
|
534
|
+
self._open_calls.add(event.call_id)
|
|
535
|
+
try:
|
|
536
|
+
arguments = event.args
|
|
537
|
+
except ValueError:
|
|
538
|
+
arguments = {}
|
|
539
|
+
await self.run_function_calls(
|
|
540
|
+
[
|
|
541
|
+
FunctionCallFromLLM(
|
|
542
|
+
function_name=event.name,
|
|
543
|
+
tool_call_id=event.call_id,
|
|
544
|
+
arguments=arguments,
|
|
545
|
+
context=self._context or LLMContext(),
|
|
546
|
+
)
|
|
547
|
+
]
|
|
548
|
+
)
|
|
549
|
+
elif isinstance(event, satham.SessionUpdated):
|
|
550
|
+
if event.language:
|
|
551
|
+
self.language = event.language
|
|
552
|
+
await self._call_event_handler("on_session_updated", event)
|
|
553
|
+
elif isinstance(event, satham.Error):
|
|
554
|
+
await self.push_error(
|
|
555
|
+
error_msg=f"Satham {event.code}: {event.message}", fatal=event.fatal
|
|
556
|
+
)
|