livekit-plugins-google 1.2.5__tar.gz → 1.2.7__tar.gz

{livekit_plugins_google-1.2.5 → livekit_plugins_google-1.2.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: livekit-plugins-google
-Version: 1.2.5
+Version: 1.2.7
 Summary: Agent Framework plugin for services from Google Cloud
 Project-URL: Documentation, https://docs.livekit.io
 Project-URL: Website, https://livekit.io/
@@ -22,7 +22,7 @@ Requires-Dist: google-auth<3,>=2
 Requires-Dist: google-cloud-speech<3,>=2
 Requires-Dist: google-cloud-texttospeech<3,>=2.27
 Requires-Dist: google-genai>=v1.23.0
-Requires-Dist: livekit-agents>=1.2.5
+Requires-Dist: livekit-agents>=1.2.7
 Description-Content-Type: text/markdown
 
 # Google AI plugin for LiveKit Agents

{livekit_plugins_google-1.2.5 → livekit_plugins_google-1.2.7}/livekit/plugins/google/beta/realtime/realtime_api.py

@@ -428,7 +428,9 @@ class RealtimeSession(llm.RealtimeSession):
         self._chat_ctx = chat_ctx.copy()
 
     async def update_tools(self, tools: list[llm.FunctionTool | llm.RawFunctionTool]) -> None:
-        new_declarations: list[types.FunctionDeclaration] = to_fnc_ctx(tools)
+        new_declarations: list[types.FunctionDeclaration] = to_fnc_ctx(
+            tools, use_parameters_json_schema=False
+        )
         current_tool_names = {f.name for f in self._gemini_declarations}
         new_tool_names = {f.name for f in new_declarations}
 
@@ -699,10 +701,15 @@ class RealtimeSession(llm.RealtimeSession):
                         break
 
                 async for response in session.receive():
-                    if (not self._current_generation or self._current_generation._done) and (
-                        response.server_content or response.tool_call
-                    ):
-                        self._start_new_generation()
+                    if not self._current_generation or self._current_generation._done:
+                        if response.server_content and response.server_content.interrupted:
+                            # interrupt a generation already done
+                            self._handle_input_speech_started()
+                            # reset the flag and still start a new generation in case it has any other content
+                            response.server_content.interrupted = False
+
+                        if self._is_new_generation(response):
+                            self._start_new_generation()
 
                     if response.session_resumption_update:
                         if (
@@ -1084,3 +1091,16 @@ class RealtimeSession(llm.RealtimeSession):
                 recoverable=recoverable,
             ),
         )
+
+    def _is_new_generation(self, resp: types.LiveServerMessage) -> bool:
+        if resp.tool_call:
+            return True
+
+        if (sc := resp.server_content) and (
+            sc.model_turn
+            or (sc.output_transcription and sc.output_transcription.text is not None)
+            or (sc.input_transcription and sc.input_transcription.text is not None)
+        ):
+            return True
+
+        return False

{livekit_plugins_google-1.2.5 → livekit_plugins_google-1.2.7}/livekit/plugins/google/llm.py

@@ -157,10 +157,6 @@ class LLM(llm.LLM):
             if _thinking_budget is not None:
                 if not isinstance(_thinking_budget, int):
                     raise ValueError("thinking_budget inside thinking_config must be an integer")
-                if not (0 <= _thinking_budget <= 24576):
-                    raise ValueError(
-                        "thinking_budget inside thinking_config must be between 0 and 24576"
-                    )
 
         self._opts = _LLMOptions(
             model=model,

{livekit_plugins_google-1.2.5 → livekit_plugins_google-1.2.7}/livekit/plugins/google/stt.py

@@ -20,6 +20,7 @@ import time
 import weakref
 from collections.abc import AsyncGenerator, AsyncIterable
 from dataclasses import dataclass
+from datetime import timedelta
 from typing import Callable, Union, cast
 
 from google.api_core.client_options import ClientOptions
@@ -28,6 +29,7 @@ from google.auth import default as gauth_default
 from google.auth.exceptions import DefaultCredentialsError
 from google.cloud.speech_v2 import SpeechAsyncClient
 from google.cloud.speech_v2.types import cloud_speech
+from google.protobuf.duration_pb2 import Duration
 from livekit import rtc
 from livekit.agents import (
     DEFAULT_API_CONNECT_OPTIONS,
@@ -67,6 +69,7 @@ class STTOptions:
     punctuate: bool
     spoken_punctuation: bool
     enable_word_time_offsets: bool
+    enable_word_confidence: bool
     model: SpeechModels | str
     sample_rate: int
     min_confidence_threshold: float
@@ -99,6 +102,7 @@ class STT(stt.STT):
         punctuate: bool = True,
         spoken_punctuation: bool = False,
         enable_word_time_offsets: bool = True,
+        enable_word_confidence: bool = False,
         model: SpeechModels | str = "latest_long",
         location: str = "global",
         sample_rate: int = 16000,
@@ -122,6 +126,7 @@ class STT(stt.STT):
             punctuate(bool): whether to punctuate the audio (default: True)
             spoken_punctuation(bool): whether to use spoken punctuation (default: False)
             enable_word_time_offsets(bool): whether to enable word time offsets (default: True)
+            enable_word_confidence(bool): whether to enable word confidence (default: False)
             model(SpeechModels): the model to use for recognition default: "latest_long"
             location(str): the location to use for recognition default: "global"
             sample_rate(int): the sample rate of the audio default: 16000
@@ -162,6 +167,7 @@ class STT(stt.STT):
             punctuate=punctuate,
             spoken_punctuation=spoken_punctuation,
             enable_word_time_offsets=enable_word_time_offsets,
+            enable_word_confidence=enable_word_confidence,
             model=model,
             sample_rate=sample_rate,
             min_confidence_threshold=min_confidence_threshold,
@@ -243,6 +249,7 @@ class STT(stt.STT):
                 enable_automatic_punctuation=config.punctuate,
                 enable_spoken_punctuation=config.spoken_punctuation,
                 enable_word_time_offsets=config.enable_word_time_offsets,
+                enable_word_confidence=config.enable_word_confidence,
             ),
             model=config.model,
             language_codes=config.languages,
@@ -547,6 +554,14 @@ class SpeechStream(stt.SpeechStream):
                 raise APIConnectionError() from e
 
 
+def _duration_to_seconds(duration: Duration | timedelta) -> float:
+    # Proto Plus may auto-convert Duration to timedelta; handle both.
+    # https://proto-plus-python.readthedocs.io/en/latest/marshal.html
+    if isinstance(duration, timedelta):
+        return duration.total_seconds()
+    return duration.seconds + duration.nanos / 1e9
+
+
 def _recognize_response_to_speech_event(
     resp: cloud_speech.RecognizeResponse,
 ) -> stt.SpeechEvent:
@@ -556,24 +571,31 @@ def _recognize_response_to_speech_event(
         text += result.alternatives[0].transcript
         confidence += result.alternatives[0].confidence
 
-    # not sure why start_offset and end_offset returns a timedelta
-    start_offset = resp.results[0].alternatives[0].words[0].start_offset
-    end_offset = resp.results[-1].alternatives[0].words[-1].end_offset
+    alternatives = []
 
-    confidence /= len(resp.results)
-    lg = resp.results[0].language_code
-    return stt.SpeechEvent(
-        type=stt.SpeechEventType.FINAL_TRANSCRIPT,
-        alternatives=[
+    # Google STT may return empty results when spoken_lang != stt_lang
+    if resp.results:
+        try:
+            start_time = _duration_to_seconds(resp.results[0].alternatives[0].words[0].start_offset)
+            end_time = _duration_to_seconds(resp.results[-1].alternatives[0].words[-1].end_offset)
+        except IndexError:
+            # When enable_word_time_offsets=False, there are no "words" to access
+            start_time = end_time = 0
+
+        confidence /= len(resp.results)
+        lg = resp.results[0].language_code
+
+        alternatives = [
             stt.SpeechData(
                 language=lg,
-                start_time=start_offset.total_seconds(),  # type: ignore
-                end_time=end_offset.total_seconds(),  # type: ignore
+                start_time=start_time,
+                end_time=end_time,
                 confidence=confidence,
                 text=text,
             )
-        ],
-    )
+        ]
+
+    return stt.SpeechEvent(type=stt.SpeechEventType.FINAL_TRANSCRIPT, alternatives=alternatives)
 
 
 def _streaming_recognize_response_to_speech_data(

{livekit_plugins_google-1.2.5 → livekit_plugins_google-1.2.7}/livekit/plugins/google/tts.py

@@ -61,6 +61,7 @@ class TTS(tts.TTS):
         language: NotGivenOr[SpeechLanguages | str] = NOT_GIVEN,
         gender: NotGivenOr[Gender | str] = NOT_GIVEN,
         voice_name: NotGivenOr[str] = NOT_GIVEN,
+        voice_cloning_key: NotGivenOr[str] = NOT_GIVEN,
         sample_rate: int = 24000,
         pitch: int = 0,
         effects_profile_id: str = "",
@@ -86,6 +87,7 @@ class TTS(tts.TTS):
             language (SpeechLanguages | str, optional): Language code (e.g., "en-US"). Default is "en-US".
             gender (Gender | str, optional): Voice gender ("male", "female", "neutral"). Default is "neutral".
             voice_name (str, optional): Specific voice name. Default is an empty string.
+            voice_cloning_key (str, optional): Voice clone key. Created via https://cloud.google.com/text-to-speech/docs/chirp3-instant-custom-voice
             sample_rate (int, optional): Audio sample rate in Hz. Default is 24000.
             location (str, optional): Location for the TTS client. Default is "global".
             pitch (float, optional): Speaking pitch, ranging from -20.0 to 20.0 semitones relative to the original pitch. Default is 0.
@@ -115,13 +117,18 @@ class TTS(tts.TTS):
 
         lang = language if is_given(language) else DEFAULT_LANGUAGE
         ssml_gender = _gender_from_str(DEFAULT_GENDER if not is_given(gender) else gender)
-        name = DEFAULT_VOICE_NAME if not is_given(voice_name) else voice_name
 
         voice_params = texttospeech.VoiceSelectionParams(
-            name=name,
             language_code=lang,
             ssml_gender=ssml_gender,
         )
+        if is_given(voice_cloning_key):
+            voice_params.voice_clone = texttospeech.VoiceCloneParams(
+                voice_cloning_key=voice_cloning_key,
+            )
+        else:
+            voice_params.name = voice_name if is_given(voice_name) else DEFAULT_VOICE_NAME
+
         if not is_given(tokenizer):
             tokenizer = tokenize.blingfire.SentenceTokenizer()
 

{livekit_plugins_google-1.2.5 → livekit_plugins_google-1.2.7}/livekit/plugins/google/utils.py

@@ -23,18 +23,27 @@ from .tools import _LLMTool
 __all__ = ["to_fnc_ctx"]
 
 
-def to_fnc_ctx(fncs: list[FunctionTool | RawFunctionTool]) -> list[types.FunctionDeclaration]:
+def to_fnc_ctx(
+    fncs: list[FunctionTool | RawFunctionTool], *, use_parameters_json_schema: bool = True
+) -> list[types.FunctionDeclaration]:
     tools: list[types.FunctionDeclaration] = []
     for fnc in fncs:
         if is_raw_function_tool(fnc):
             info = get_raw_function_info(fnc)
-            tools.append(
-                types.FunctionDeclaration(
-                    name=info.name,
-                    description=info.raw_schema.get("description", ""),
-                    parameters_json_schema=info.raw_schema.get("parameters", {}),
+            fnc_kwargs = {
+                "name": info.name,
+                "description": info.raw_schema.get("description", ""),
+            }
+            if use_parameters_json_schema:
+                fnc_kwargs["parameters_json_schema"] = info.raw_schema.get("parameters", {})
+            else:
+                # https://github.com/googleapis/python-genai/issues/1147
+                fnc_kwargs["parameters"] = types.Schema.from_json_schema(
+                    json_schema=types.JSONSchema.model_validate(
+                        info.raw_schema.get("parameters", {})
+                    )
                 )
-            )
+            tools.append(types.FunctionDeclaration(**fnc_kwargs))
 
         elif is_function_tool(fnc):
             tools.append(_build_gemini_fnc(fnc))

{livekit_plugins_google-1.2.5 → livekit_plugins_google-1.2.7}/livekit/plugins/google/version.py

@@ -12,4 +12,4 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-__version__ = "1.2.5"
+__version__ = "1.2.7"

{livekit_plugins_google-1.2.5 → livekit_plugins_google-1.2.7}/pyproject.toml

@@ -27,7 +27,7 @@ dependencies = [
     "google-cloud-speech >= 2, < 3",
     "google-cloud-texttospeech >= 2.27, < 3",
     "google-genai >= v1.23.0",
-    "livekit-agents>=1.2.5",
+    "livekit-agents>=1.2.7",
 ]
 
 [project.urls]