docs: add Voice Modules section for consolidated voice/ tree
Browse files
README.md
CHANGED
|
@@ -24,4 +24,21 @@ APOLLO is the required optimizer for later fine-tuned releases. It keeps optimiz
|
|
| 24 |
|
| 25 |
Quantization policy: Q4_K_M is the default published runtime artifact today; Q6_K and Q8_0 are tracked in the app catalog as higher-precision variants for the hardware optimizer to select when those files are published.
|
| 26 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
Repository: `elizaos/eliza-1`
|
|
|
|
| 24 |
|
| 25 |
Quantization policy: Q4_K_M is the default published runtime artifact today; Q6_K and Q8_0 are tracked in the app catalog as higher-precision variants for the hardware optimizer to select when those files are published.
|
| 26 |
|
| 27 |
+
## Voice Modules
|
| 28 |
+
|
| 29 |
+
The voice sub-models live under `voice/<sub>/`. The unified layout
|
| 30 |
+
replaces the legacy `elizaos/eliza-1-voice-*` per-component repos
|
| 31 |
+
(consolidated 2026-05-15). Each sub-path carries a `provenance.txt`
|
| 32 |
+
linking back to its source repo at the time of consolidation.
|
| 33 |
+
|
| 34 |
+
| Sub-path | Contents | Source repo (deleted) |
|
| 35 |
+
| --- | --- | --- |
|
| 36 |
+
| `voice/omnivoice/` | OmniVoice TTS — `omnivoice-base-q4_k_m.gguf`, `omnivoice-base-q8_0.gguf`, `omnivoice-tokenizer-q4_k_m.gguf`, presets including `voice-preset-same.bin` and `voice-preset.elz2` (same-voice v0.1) plus README/eval/manifest | `elizaos/eliza-1-voice-omnivoice-same-v01` |
|
| 37 |
+
| `voice/emotion/` | Wav2Small emotion classifier — `wav2small-cls7-int8.onnx` (cls7 head) plus its config JSON, `eval.json`, `manifest.json`, README | `elizaos/eliza-1-voice-emotion` |
|
| 38 |
+
| `voice/turn/` | LiveKit turn detector — `intl/turn-detector-intl-q8.gguf` (281 MB), `intl/model_q8.onnx` (262 MB), tokenizer assets, README, manifest | `elizaos/eliza-1-voice-turn` |
|
| 39 |
+
| `voice/asr/` | Reserved (legacy repo was empty); ASR weights live under the canonical `voice/asr/` path: `eliza-1-asr-q8_0.gguf`, `eliza-1-asr-mmproj.gguf` | `elizaos/eliza-1-voice-asr` |
|
| 40 |
+
| `voice/kokoro/` | Kokoro TTS — extra preset `voices/af_sam.bin` (the canonical Kokoro weights live alongside under `voice/kokoro/kokoro-v1.0-q4.onnx` and `voices/af_bella.bin`, `voices/af_same.bin`) | `elizaos/eliza-1-voice-kokoro` |
|
| 41 |
+
|
| 42 |
+
Other voice components (`voice/diarizer/`, `voice/wakeword/`, `voice/speaker-encoder/`, `voice/vad/`, `voice/embedding/`, `voice/turn-detector/`, `voice/voice-emotion/`) were already shipping under the unified repo and remain unchanged.
|
| 43 |
+
|
| 44 |
Repository: `elizaos/eliza-1`
|