ovos-tts-plugin-pocket-tts
OVOS TTS plugin for Pocket TTS by Kyutai — a lightweight, CPU-only text-to-speech engine with a 100M parameter model.
~6x real-time on Apple Silicon, no GPU required. Supports voice cloning from audio samples.
Install
pip install ovos-tts-plugin-pocket-tts
With uv, torch automatically resolves to the CPU-only wheel (configured via tool.uv.sources).
Linux: CPU-only torch (saves ~2GB)
On Linux, pip defaults to the CUDA torch wheel (~2.5GB). If you don't need GPU support (pocket-tts is CPU-only anyway), install torch from the CPU index first:
pip install torch --index-url https://download.pytorch.org/whl/cpu
pip install ovos-tts-plugin-pocket-tts
On macOS, this is not needed — PyPI torch is already CPU-only (~60MB).
Configuration
{
"tts": {
"module": "ovos-tts-plugin-pocket-tts",
"ovos-tts-plugin-pocket-tts": {
"voice": "alba"
}
}
}
Voice options
Built-in voices: alba, marius, javert, jean, fantine, cosette, eponine, azelma
Voice cloning: Set voice to a path to a .wav file, a pre-exported .safetensors file, or a HuggingFace URI (hf://kyutai/tts-voices/...).
Language support
This plugin auto-detects which languages your installed pocket-tts supports.
- Legacy
pocket-tts(≤ 1.1.1, current PyPI release): English only. - Multilingual
pocket-tts(kyutai-labs/pocket-tts#155, not yet released): English, French, German, Spanish, Italian, Portuguese.
When the multilingual release lands on PyPI, just pip install -U pocket-tts and the plugin lights up the new languages automatically — no plugin upgrade required.
How language selection works
The plugin maps the active OVOS language (BCP-47, e.g. fr-FR) to a kyutai pocket-tts model id. Defaults:
| OVOS lang | Kyutai model |
|---|---|
en |
english_v2 |
fr |
french_24l |
de |
german_24l |
es |
spanish_24l |
it |
italian_24l |
pt |
portuguese_24l |
Lookup tries the full BCP-47 tag first (e.g. pt-br), then falls back to the base subtag (pt), then to English. Unknown languages fall back to English with a log line.
Note: The
_24l(24-layer) models are larger preview builds, not the final distilled releases. They are slower than the english defaults but support int8 quantization for a meaningful speedup. The plugin enablesquantize=Trueautomatically for any*_24lmodel. You can override per-language via thequantizeconfig knob.
Override the language map
If you want, for example, the older english_v1 model, or to register a future regional model, set language_aliases in your config:
{
"tts": {
"module": "ovos-tts-plugin-pocket-tts",
"ovos-tts-plugin-pocket-tts": {
"voice": "alba",
"language_aliases": {
"en": "english_v1",
"pt-br": "portuguese_brazil_24l"
},
"quantize": {
"french_24l": false
},
"preload_languages": ["en", "fr"]
}
}
}
| Key | Type | Default | Description |
|---|---|---|---|
voice |
str | "alba" |
Built-in voice name, .wav/.safetensors path, or hf:// URI. |
sample_rate |
int | 16000 |
Output sample rate in Hz. The model is resampled if it differs. |
language_aliases |
dict | {} |
Override or extend the BCP-47 → kyutai model map. Full tags take precedence over base subtags. |
quantize |
bool or dict | auto (True for *_24l) |
Force int8 quantization on/off. Pass a dict to control per kyutai model id. |
preload_languages |
list[str] | [] |
BCP-47 codes to load eagerly during plugin init instead of lazy-loading on first use. |
enable_streaming |
bool | false |
Use the streaming TTS path (recommended for low-latency setups). |
Memory note: Each loaded language model holds ~100M parameters in RAM (the
*_24lpreviews are ~4× bigger before distillation). The plugin caches one model per used language, so leavingpreload_languagesempty and letting the cache warm on demand keeps the resident set small.
Roadmap
- SouraTTS (emotional TTS built on pocket-tts) is being tracked as a separate plugin (
ovos-tts-plugin-soura-tts) — its emotion/intensity dimensions don't fit the current OVOS TTS interface and bundling would tie pocket-tts upgrades to SouraTTS releases.
License
Apache-2.0
Metadata
Release files for ovos-tts-plugin-pocket-tts 0.2.0
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 | Actions |
|---|---|---|---|---|
| ovos_tts_plugin_pocket_tts-0.2.0.tar.gz | 18.2 kB | Without provenance | Details for ovos_tts_plugin_pocket_tts-0.2.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| ovos_tts_plugin_pocket_tts-0.2.0-py3-none-any.whl | Python 3 | none | any | Details for ovos_tts_plugin_pocket_tts-0.2.0-py3-none-any.whl |
Total release size: 32.0 kB
Release files / ovos_tts_plugin_pocket_tts-0.2.0.tar.gz
| Download URL | ovos_tts_plugin_pocket_tts-0.2.0.tar.gz |
|---|---|
| Size | 18.2 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
c368881058ef9626d3c4ad69ddbf38f846ebd2aeabdf7f863f2ac01e90fe302b
|
|
BLAKE2b-256 checksum How to use checksums |
3af153c7427b2bf9cc02b7a5e48bfedbdeac4c8757abaf751a6383442521e3c8
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Release files / ovos_tts_plugin_pocket_tts-0.2.0-py3-none-any.whl
| Download URL | ovos_tts_plugin_pocket_tts-0.2.0-py3-none-any.whl |
|---|---|
| Size | 13.8 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
c83a5ee81efd28af465c5b13d99e721d8aeafb9f764c4824f7947761bd501ae0
|
|
BLAKE2b-256 checksum How to use checksums |
0dcb4ed0a0816581bb55052f3c605759eb24ac8e639454cbfd7949e47ac68cb1
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Security for ovos-tts-plugin-pocket-tts 0.2.0
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.