ema-lightning 1.0.0__tar.gz

ema_lightning-1.0.0/.gitignore

@@ -0,0 +1,9 @@
+__pycache__/
+*.py[cod]
+.pytest_cache/
+.ruff_cache/
+build/
+dist/
+*.egg-info/
+*.wav
+.DS_Store

ema_lightning-1.0.0/CHANGELOG.md

@@ -0,0 +1,21 @@
+# Changelog
+
+## 1.0.0 (unreleased)
+
+First release.
+
+- `EMA()` with `say()` for one text or a list, `stream()` and `best_batch_size()`.
+- `lightning()`: cuDNN benchmark mode, one compilation for any batch size, and CUDA
+  graphs at batch sizes 1, 2, 4 and every multiple of 8 up to the batch size,
+  checked against the plain path at startup.
+- Text frontend on [normalizer-tr](https://github.com/erdemtuna/normalizer-tr)
+  0.4 with the `fallback` policy: numbers, dates, times, money, units,
+  abbreviations and symbols are read aloud, and nothing is skipped.
+- Playhead, a scheduler inside `EMA`: every `say()` and `stream()` call joins two
+  first-come-first-served queues (sentences before the model, windows before the
+  decoder) that run in shared GPU batches, with no server or database. Closed
+  streams leave both queues, and a failing batch fails only its own callers.
+  `benches/callers.py` measures it under load.
+- Decoder windows of four seconds; a stream's first window is one second. On a
+  GPU, windows are padded to 48 or 120 frames and each decode batch holds one size.
+- Requires CPython 3.11 or newer, the minimum of normalizer-tr 0.4.

ema_lightning-1.0.0/CONTRIBUTING.md

@@ -0,0 +1,96 @@
+# Contributing
+
+Report bugs with a minimal example: the text, the settings (`speed`, `seed`,
+`sample_rate`), what you expected to hear, what you heard, the package version,
+PyTorch version and device. Attach audio only if you made it with a fixed seed.
+Do not include private or customer text.
+
+If a number, date, abbreviation or symbol is read wrongly, check it with
+normalizer-tr directly first:
+
+```python
+from normalizer_tr import Normalizer
+print(Normalizer().normalize("3. kat", ambiguity_policy="fallback").normalized_text)
+```
+
+If normalizer-tr gives the same reading, report it
+[there](https://github.com/erdemtuna/normalizer-tr/issues).
+
+## Set up
+
+```sh
+git clone <this repository>
+cd ema-lightning
+python -m venv .venv && . .venv/bin/activate
+python -m pip install -e ".[dev]"
+python -m pytest -q
+```
+
+The tests use tiny random models with the real architecture, so they need no
+weights, no network and no GPU. GPU tests run only when CUDA is available.
+
+## Code layout
+
+| Module | Owns |
+|---|---|
+| `api.py` | The public `EMA` and `Speech`, settings validation, seeds and WAV output |
+| `frontend.py` | Cleaning, normalizer-tr with the fallback policy, the model's alphabet |
+| `chunker.py` | Splitting text into pieces with pauses |
+| `scheduler.py` | Playhead: the background loop that owns the GPU, with the two first-come-first-served queues (sentences, then windows) |
+| `engine.py` | The three stages Playhead runs: plan (durations and the frame timeline), think (sampling) and decode (windows, padded to the decoder sizes on a GPU) |
+| `model.py` | The acoustic model: text encoder, durations, aligner and latent generator |
+| `decoder.py` | The decoder from latents to 48 kHz audio |
+| `graphs.py` | Compilation and CUDA graphs for `lightning()`: the batch-size ladder and the length buckets |
+| `audio.py` | Resampling and WAV writing |
+
+Language rules belong in normalizer-tr, not here. `frontend.py` only cleans, calls
+it and maps to the alphabet.
+
+## Rules
+
+- Text never raises. A failing normalizer keeps the text as written.
+- The same audio made two ways (a list, a stream, many threads, the fast path)
+  must match saying the text alone: the same length, with the difference at least
+  40 dB quieter than the speech. `same_speech()` in `tests/conftest.py` checks it.
+- Only Playhead's loop runs the engine during `say()` and `stream()`. Anything
+  else that uses the engine directly (startup checks, measurements) holds
+  `engine.lock`.
+- Both queues are strictly first come, first served. Whatever does not fit in a
+  batch stays at the front for the next turn, and nothing is ever reordered. On a
+  GPU, a decode batch holds one window size.
+- A failing batch fails only the callers in it; the loop keeps serving.
+- The compiled path must match the plain path; `lightning()` checks this at startup.
+- Invalid settings raise `ValueError` before any work.
+- Add a test with every change in behavior. Do not change expected readings just
+  to make a test pass.
+
+## Checks
+
+```sh
+python -m ruff check src tests examples benches
+python -m pytest -q
+```
+
+| Tests | What they check |
+|---|---|
+| `test_api.py` | The public calls: `say()`, `stream()`, sample rates, seeds, lists, empty text, bad settings, files, leaving a stream early |
+| `test_playhead.py` | Both queues' order, overflow, one window size per batch, hang-ups, a failure in each stage, many threads. Runs on a recording stand-in engine, so batch contents are checked exactly |
+| `test_engine.py` | Windows, padding to the decoder sizes, the batch-size ladder, the bucketed path against the plain path |
+| `test_model.py` | The model and the decoder are exact under padding |
+| `test_chunker.py` | Cutting text into sentences and clauses, with pauses |
+| `test_frontend.py` | Cleaning, normalizer-tr with the fallback policy, the alphabet |
+| `test_audio.py` | Resampling and WAV writing |
+| `test_gpu.py` | `lightning()` on a real GPU: graphs against the plain path, streams and lists, benchmark mode. Skipped without CUDA |
+
+On a GPU, run `benches/callers.py` after any change to `scheduler.py`.
+
+CI runs both on CPython 3.11, 3.12 and 3.13 without weights or a GPU.
+
+## Releases
+
+Bump the version in `pyproject.toml` and `src/ema_lightning/__init__.py`, add the
+changes to `CHANGELOG.md`, re-measure [PERFORMANCE.md](PERFORMANCE.md) if the model
+or frontend changed, then build and publish. Publishing is done by the owner only.
+
+Contributions are Apache-2.0. Keep weights, audio, environments and credentials
+out of Git.

ema_lightning-1.0.0/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

ema_lightning-1.0.0/PERFORMANCE.md

@@ -0,0 +1,69 @@
+# Performance and measurement
+
+Measured on the release candidate (5.6M-parameter acoustic model with the
+3M-parameter decoder) at speed 1.0. Accuracy was measured with the release
+candidate's own text normalizer, before the switch to normalizer-tr.
+
+## Accuracy
+
+Freya-TR-Eval, all 495 sentences, three seeds, raw text in.
+
+| Metric | Value |
+|---|---|
+| Word error rate | 0.92% (0.79 to 1.05 across seeds) |
+| Character error rate | 0.18% |
+
+Transcripts come from Whisper large-v3 (beam size 5) on an 8 kHz band, scored
+with Freya's text normalization. Some counted errors are the recognizer's.
+
+## Speed
+
+RTX 4090. Real-time factor (RTF) is generation time divided by audio length:
+three passes after a warm-up, middle pass reported.
+
+| | Plain PyTorch | `lightning()` |
+|---|---|---|
+| One request | RTF 0.0115 (87× real time) | RTF 0.0023 (440× real time) |
+| Batch of 64 | 952× real time | 1,316× real time |
+| First audio, typical sentence | | 3.86 ms, normalization included |
+
+On a CPU the model runs at RTF 0.166, about 6× real time (measured on a
+cloud-container CPU).
+
+At $0.74 an hour for a rented RTX 4090, the batched compiled path reads about
+24,000 characters a second: about $0.0085 per million characters.
+
+## The package on an RTX PRO 6000
+
+Measured with this package (`lightning()`, cuDNN benchmark mode on), RTX PRO 6000
+Blackwell Server Edition, PyTorch 2.11.
+
+| | |
+|---|---|
+| First audio from `stream()` | 4.3 ms typical (4.2 to 4.4 ms with graphs recorded up to batch 8, 32, 64 or 128) |
+| 1,000 short texts in one `say()` list | about 1,350× real time |
+| Plain path, best batch | 1,431× real time at batch 16 |
+
+## Many callers
+
+`benches/callers.py` starts 1, 8, 32, 64 and 128 callers at the same instant, each
+calling `say()` with one sentence, and reports how long the calls took (median and
+95th percentile) and the total real-time multiple.
+
+```sh
+python -m pip install tqdm
+python benches/callers.py
+```
+
+## Measure on your machine
+
+```python
+import time
+from ema_lightning import EMA
+
+tts = EMA().lightning()                     # prints first audio, a short sentence's time and the batch size
+text = "Bugün hava çok güzel, yarın da yağmur yağacakmış."
+start = time.perf_counter()
+speech = tts.say(text, seed=0)
+print("RTF", (time.perf_counter() - start) / speech.duration)
+```