voicesynth
Offline voice synthesis libraryInstall
pip install voicesynth
pip install torch --index-url https://download.pytorch.org/whl/cpu
PyTorch is not pulled in automatically so you can pick the build you want (CPU, CUDA, ...). pip install voicesynth[torch] installs the default one.
Quick start
from voicesynth import Model, Synthesizer
synthesizer = Synthesizer(Model("v5_6_ru"))
synthesizer.say("В н+едрах т+ундры в+ыдры в г+етрах т+ырят в в+ёдра +ядра к+едров.")
The model is downloaded on first use into ~/.cache/voicesynth (override with model_path= or the VOICESYNTH_HOME environment variable).
Models
from voicesynth import show_available_models, available_models
show_available_models()
show_available_models("tatar")
available_models("russian")
| Family | Models |
|---|---|
| Russian | v5_6_ru, v5_5_ru, v5_4_ru, v5_3_ru, v5_2_ru, v5_1_ru, v5_ru, v4_ru, v3_1_ru |
| CIS languages | v5_cis_base, v5_cis_base_nostress, v5_cis_ext |
| Turkic / Caucasian | v5_turkic, v5_caucasian |
| Cyrillic | v4_cyrillic |
| Ukrainian / Uzbek / Tatar / Kalmyk | v4_ua, v3_ua, v4_uz, v3_uz, v3_tt, v3_xal |
| Indic | v4_indic, v3_indic, v3_en_indic |
| English / German / Spanish / French | v3_en, v3_de, v3_es, v3_fr |
v5_cis_base, v5_cis_ext, v5_turkic and v5_caucasian expect stress marks (к+ошка) in the text; the nostress variant needs them only for Slavic languages. Speakers are prefixed with the language code, for example tat_albina or kaz_abai.
Speakers and speed
model = Model("v3_en", speaker="en_73")
model.set_speaker("en_12")
synthesizer = Synthesizer(model)
synthesizer.say("Hello there", speaker="en_5", prosody_rate=90)
print(synthesizer.speakers)
Saving audio and other outputs
synthesizer.synthesize("Saved for later", path="audio.wav")
synthesizer.say("Saved and played", path="audio.wav")
wav = synthesizer.synthesize_bytes("In memory")
samples = synthesizer.generate("As a numpy array")
synthesizer.synthesize("Into a buffer", io.BytesIO())
say() without a path plays through a temporary file. Pass ssml=True to send your own SSML, sample_rate= (8000, 24000 or 48000) to change quality, and device="cuda" or threads=4 to the Synthesizer constructor.
Long text
Text longer than max_chars (500 by default) is split on sentence and clause boundaries, synthesized piece by piece and joined with pause seconds of silence:
synthesizer = Synthesizer(model, max_chars=300, pause=0.4)
from voicesynth import split_text
split_text(long_text, 300)
Non-blocking playback and asyncio
playback = synthesizer.say("Talking in the background", blocking=False)
playback.is_playing
playback.wait(timeout=2)
playback.stop()
await synthesizer.say_async("Does not block the event loop")
await synthesizer.synthesize_async("hello", "hello.wav")
Audio playback is native: winsound on Windows, afplay on macOS and the first available of paplay, pw-play, aplay, ffplay, mpv, play on Linux.
Speakers
Some catalog entries (Indic, German, Tatar, Kalmyk and a few more) list only the speakers known in advance. After loading a model, synthesizer.speakers holds the real list. Without creating a synthesizer:
Model("v3_de").fetch_speakers()
Model cache
Models are stored in ~/.cache/voicesynth (or $VOICESYNTH_HOME). Every download is checked: the size, the archive CRCs and a SHA-256 checksum saved next to the file, which is re-verified on every Model(...). A damaged file is deleted and downloaded again. Interrupted downloads resume where they stopped.
from voicesynth import cache
cache.download("v3_en")
cache.cached_models()
cache.verify("v3_en")
cache.remove("v3_en")
cache.clear()
Mixed language text
Cyrillic models read Cyrillic and Latin models read Latin letters. When text contains words in the other script, a warning is logged because those words are skipped or mispronounced. Silence it with Synthesizer(model, warn_mixed_language=False).
Command line
voicesynth models
voicesynth models --language russian
voicesynth speakers v3_en
voicesynth speakers v3_de --load
voicesynth say "Hello world" -m v3_en -s en_73
voicesynth say "Saved only" -m v3_en -o hello.wav --no-play
echo "Piped text" | voicesynth say -m v3_en
voicesynth cache list
voicesynth cache download v3_en v5_6_ru
voicesynth cache verify
voicesynth cache remove v3_en
voicesynth cache clear
Logging
from voicesynth import disable_logging, enable_logging
disable_logging()
Errors
All exceptions derive from voicesynth.exceptions.VoiceSynthError: IncorrectModel, IncorrectSpeaker, ModelDownloadError, SynthesisError, AudioPlayingError, ModelIntegrityError, PackageError and others.
Tests
pip install voicesynth[dev]
pytest
mypy
1 maintainer
Metadata
1 maintainer
Release files for voicesynth 0.5.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | |
|---|---|---|---|---|
| voicesynth-0.5.0.tar.gz | 30.8 kB | Without provenance | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| voicesynth-0.5.0-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 55.1 kB
Release files / voicesynth-0.5.0.tar.gz
| Download URL | voicesynth-0.5.0.tar.gz |
|---|---|
| Size | 30.8 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
848bfa56a0a74bfbeced3b22e0bc5a04fc045176433f46cd589672880f253ef2
|
|
BLAKE2b-256 checksum How to use checksums |
14a34944e1569b9452d98f90a32d8107d498aa4d22edd03ac8ead3aaa53ebb65
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.12.14
|
Release files / voicesynth-0.5.0-py3-none-any.whl
| Download URL | voicesynth-0.5.0-py3-none-any.whl |
|---|---|
| Size | 24.3 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
7db43867a42ace388533aac24824eceb97f21625d9de8a91e48dcc35ba2e1381
|
|
BLAKE2b-256 checksum How to use checksums |
d892af922534f7bcbe3c4a4be1b2a46e9b6583c71fcc1b033a856c2a071d897b
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.12.14
|
Security for voicesynth 0.5.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.
