This release is a pre-release and may not be stable for production use.
ovos-voice-embeddings-plugin
An OVOS voice embeddings plugin. It turns a speech clip into a speaker
embedding with speakeronnx, so
the runtime needs onnxruntime and numpy only, no PyTorch. The default model
is wespeaker-resnet34, a 256-dimension L2-normalised embedding downloaded from
the OpenVoiceOS speaker-embeddings-onnx
collection on first use. Two clips of one speaker give vectors with a high
cosine similarity. Clips of different speakers give a low one.
The plugin implements the VoiceEmbedder template of
ovos-plugin-manager and
registers under the opm.embeddings.voice entry point as
ovos-voice-embeddings-plugin.
Install
pip install ovos-voice-embeddings-plugin
Usage
import numpy as np
import soundfile as sf
from ovos_voice_embeddings import SpeakerOnnxVoiceEmbedder
embedder = SpeakerOnnxVoiceEmbedder() # wespeaker-resnet34, 16 kHz input
alice_1, _ = sf.read("alice_1.wav", dtype="float32")
alice_2, _ = sf.read("alice_2.wav", dtype="float32")
bob, _ = sf.read("bob.wav", dtype="float32")
a1 = embedder.get_embeddings(alice_1)
a2 = embedder.get_embeddings(alice_2)
b = embedder.get_embeddings(bob)
print(a1.shape) # (256,)
print(float(np.dot(a1, a2))) # same speaker, about 0.8
print(float(np.dot(a1, b))) # different speakers, about 0.4
The embeddings are L2-normalised, so the dot product is the cosine similarity.
get_embeddings also accepts raw signed 16-bit PCM bytes, which is what the
OVOS listener hands to plugins.
Store and match voices
Pair the embedder with any OVOS embeddings database plugin, for example ovos-chromadb-embeddings-plugin:
from ovos_chromadb_embeddings import ChromaEmbeddingsDB
db = ChromaEmbeddingsDB({"path": "./voice_db"})
db.add_embeddings("alice", a1)
db.add_embeddings("bob", b)
print(db.query(a2, top_k=1)) # [("alice", distance)]
Configuration
| key | default | description |
|---|---|---|
model |
"wespeaker-resnet34" |
speakeronnx model alias or path to an ONNX file |
sample_rate |
16000 |
sample rate of the audio passed to get_embeddings |
Tests
pip install -e ".[test]"
pytest tests
The tests embed three real clips from LibriSpeech test-clean (CC BY 4.0) and assert the vector shape and that two clips of one speaker are closer than clips of two speakers.
License
Apache-2.0.
Metadata
Release files for ovos-voice-embeddings-plugin 0.0.1a1
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_voice_embeddings_plugin-0.0.1a1.tar.gz | 8.2 kB | Without provenance | Details for ovos_voice_embeddings_plugin-0.0.1a1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| ovos_voice_embeddings_plugin-0.0.1a1-py3-none-any.whl | Python 3 | none | any | Details for ovos_voice_embeddings_plugin-0.0.1a1-py3-none-any.whl |
Total release size: 16.6 kB
Release files / ovos_voice_embeddings_plugin-0.0.1a1.tar.gz
| Download URL | ovos_voice_embeddings_plugin-0.0.1a1.tar.gz |
|---|---|
| Size | 8.2 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
cf92d080cc21f41d448a6b9618d0aaa3fdebb7e12bd6bd30cc74a69c0424629b
|
|
BLAKE2b-256 checksum How to use checksums |
2aa132eb912c934c20d756946828c5689181bb4d40da2505434e4f189bf5ef44
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Release files / ovos_voice_embeddings_plugin-0.0.1a1-py3-none-any.whl
| Download URL | ovos_voice_embeddings_plugin-0.0.1a1-py3-none-any.whl |
|---|---|
| Size | 8.4 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
09c3d84aab89a8b97d1903001d65f7db0bea19a1a6cd569d2de89fa841e86f17
|
|
BLAKE2b-256 checksum How to use checksums |
0b38cefb182e307ae6916ebc1cfb6cdcd9186da966dfa67b97c16e2ba1179d3e
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Security for ovos-voice-embeddings-plugin 0.0.1a1
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.