speakeronnx 0.0.1__tar.gz

speakeronnx-0.0.1/PKG-INFO

@@ -0,0 +1,163 @@
+Metadata-Version: 2.4
+Name: speakeronnx
+Version: 0.0.1
+Summary: Pure-onnxruntime speaker embedding library — no torch at runtime
+Author-email: JarbasAI <jarbasai@mailfence.com>
+License: Apache-2.0
+Project-URL: Homepage, https://github.com/TigreGotico/speakeronnx
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Operating System :: OS Independent
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+Requires-Dist: numpy
+Requires-Dist: onnxruntime
+Requires-Dist: huggingface_hub
+Provides-Extra: soxr
+Requires-Dist: soxr; extra == "soxr"
+Provides-Extra: test
+Requires-Dist: pytest; extra == "test"
+Requires-Dist: edge-tts; extra == "test"
+Requires-Dist: requests; extra == "test"
+
+# speakeronnx
+
+Pure-onnxruntime speaker embedding library — no torch at runtime.
+
+Extract speaker embeddings, compute cosine similarity, and verify speaker identity
+using ONNX-exported models downloaded automatically from HuggingFace.
+
+**Model collection:** [OpenVoiceOS/speaker-embeddings-onnx](https://huggingface.co/collections/OpenVoiceOS/speaker-embeddings-onnx)
+
+## Install
+
+```bash
+pip install speakeronnx
+```
+
+Optional high-quality resampling:
+```bash
+pip install speakeronnx soxr
+```
+
+## Quick start
+
+```python
+from speakeronnx import SpeakerEmbedder, cosine, verify
+
+embedder = SpeakerEmbedder(model="wespeaker-resnet34")
+
+alice1 = embedder.embed("alice_clip1.wav")
+alice2 = embedder.embed("alice_clip2.wav")
+bob    = embedder.embed("bob_clip1.wav")
+
+print(cosine(alice1, alice2))   # e.g. 0.82  — same speaker
+print(cosine(alice1, bob))      # e.g. 0.21  — different speaker
+
+ok, score = verify(alice1, alice2, threshold=0.45)
+print(ok, score)  # True 0.82
+```
+
+More examples in [`examples/`](examples/).
+
+## CLI
+
+```bash
+speakeronnx list                              # list available models
+speakeronnx embed clip.wav                    # extract embedding
+speakeronnx verify a.wav b.wav               # same-speaker check (exit 0/1)
+speakeronnx verify a.wav b.wav --threshold 0.5
+speakeronnx embed clip.wav --model wespeaker-ecapa512
+```
+
+Full CLI reference in [`docs/cli.md`](docs/cli.md).
+
+## Models
+
+All 9 models are registered in `MODEL_REGISTRY` and downloaded on first use:
+
+| Alias | Embed dim | Frontend | License |
+|---|---|---|---|
+| `wespeaker-resnet34` | 256 | fbank80 | cc-by-4.0 |
+| `wespeaker-ecapa512` | 192 | fbank80 | cc-by-4.0 |
+| `wespeaker-resnet293` | 256 | fbank80 | cc-by-4.0 |
+| `campplus` | 512 | fbank80 | cc-by-4.0 |
+| `campplus-zh-en` | 192 | fbank80 | apache-2.0 |
+| `eres2net` | 192 | fbank80 | apache-2.0 |
+| `titanet-small` | 192 | fbank80 | cc-by-4.0 |
+| `titanet-large` | 192 | fbank80 | cc-by-4.0 |
+| `redimnet-b2` | 192 | raw | apache-2.0 |
+
+Full model comparison and selection guide in [`docs/models.md`](docs/models.md).
+
+## Documentation
+
+| Document | Description |
+|---|---|
+| [`docs/index.md`](docs/index.md) | Full getting-started guide |
+| [`docs/models.md`](docs/models.md) | Model comparison, selection, frontend/layout details |
+| [`docs/api.md`](docs/api.md) | Complete API reference |
+| [`docs/cli.md`](docs/cli.md) | CLI usage reference |
+| [`docs/frontend.md`](docs/frontend.md) | Feature frontend (fbank80 vs raw) technical details |
+| [`docs/advanced.md`](docs/advanced.md) | Custom models, GPU, threshold tuning |
+
+## Examples
+
+| Script | Description |
+|---|---|
+| [`examples/basic_embedding.py`](examples/basic_embedding.py) | Extract embedding from a single WAV |
+| [`examples/verify_speakers.py`](examples/verify_speakers.py) | Verify two clips, try multiple thresholds |
+| [`examples/compare_models.py`](examples/compare_models.py) | Compare all models on same utterances |
+| [`examples/batch_enrollment.py`](examples/batch_enrollment.py) | Enroll speakers from directories, match unknown |
+| [`examples/custom_model.py`](examples/custom_model.py) | Load a custom ONNX model from disk |
+| [`examples/gpu_inference.py`](examples/gpu_inference.py) | CUDA / CoreML inference |
+
+## Tests
+
+```bash
+# Unit tests (mocked, no downloads, no network)
+pytest tests/test_unit.py tests/test_audio.py tests/test_frontend.py \
+      tests/test_embedder.py tests/test_cli.py tests/test_model_registry.py -v
+| Alias | HF repo | License | Embed dim | Description |
+|---|---|---|---|---|
+| `wespeaker-resnet34` | [Wespeaker/wespeaker-voxceleb-resnet34-LM](https://huggingface.co/Wespeaker/wespeaker-voxceleb-resnet34-LM) | cc-by-4.0 | 256 | ResNet34 r-vector, VoxCeleb2 Dev — **recommended default** |
+| `wespeaker-ecapa512` | [Wespeaker/wespeaker-ecapa-tdnn512-LM](https://huggingface.co/Wespeaker/wespeaker-ecapa-tdnn512-LM) | cc-by-4.0 | 192 | ECAPA-TDNN-512 x-vector, VoxCeleb2 Dev |
+| `wespeaker-resnet293` | [Wespeaker/wespeaker-voxceleb-resnet293-LM](https://huggingface.co/Wespeaker/wespeaker-voxceleb-resnet293-LM) | cc-by-4.0 | 256 | ResNet293 r-vector — highest accuracy, 28M params |
+| `campplus` | [csukuangfj/speaker-embedding-models](https://huggingface.co/csukuangfj/speaker-embedding-models) | cc-by-4.0 | 512 | CAM++ (D-TDNN backbone), VoxCeleb2 Dev |
+| `campplus-zh-en` | [csukuangfj/speaker-embedding-models](https://huggingface.co/csukuangfj/speaker-embedding-models) | apache-2.0 | 192 | 3D-Speaker CAM++ multilingual (zh+en) |
+| `eres2net` | [csukuangfj/speaker-embedding-models](https://huggingface.co/csukuangfj/speaker-embedding-models) | apache-2.0 | 192 | ERes2Net, VoxCeleb |
+| `titanet-small` | [csukuangfj/speaker-embedding-models](https://huggingface.co/csukuangfj/speaker-embedding-models) | cc-by-4.0 | 192 | NVIDIA NeMo TitaNet-small (~40 MB) |
+| `titanet-large` | [csukuangfj/speaker-embedding-models](https://huggingface.co/csukuangfj/speaker-embedding-models) | cc-by-4.0 | 192 | NVIDIA NeMo TitaNet-large (~101 MB) |
+| `redimnet-b2` | [OpenVoiceOS/redimnet-b2-vox2-onnx](https://huggingface.co/OpenVoiceOS/redimnet-b2-vox2-onnx) | apache-2.0 | 192 | ReDimNet b2 (1.8M params), raw audio input |
+
+# End-to-end tests (downloads models + generates TTS audio)
+pytest tests/test_e2e.py -v -s
+```
+
+Use `speakeronnx list` to print descriptions and metadata for all registered models.
+
+## Feature frontend
+
+- **fbank80** models: 80-dim log-Mel filterbank with per-utterance CMN,
+  implemented in pure numpy. See [`docs/frontend.md`](docs/frontend.md).
+- **raw** models (redimnet-b2): raw 16 kHz waveform passed directly
+  to ONNX (internal MelSpectrogram in the model).
+
+## Audio requirements
+
+- Mono PCM WAV, any bit depth (8/16/24/32-bit int, 32-bit float)
+- Any sample rate (resampled internally to 16 kHz)
+- Stereo files are downmixed to mono
+- Minimum ~1 second; recommended enrollment 5–30 seconds per speaker
+
+## Dependencies
+
+- `onnxruntime`
+- `numpy`
+- `huggingface_hub`
+- `soxr` (optional, for high-quality resampling)
+
+## Project links
+
+- **GitHub:** [TigreGotico/speakeronnx](https://github.com/TigreGotico/speakeronnx)
+- **PyPI:** `pip install speakeronnx`

speakeronnx-0.0.1/README.md

@@ -0,0 +1,141 @@
+# speakeronnx
+
+Pure-onnxruntime speaker embedding library — no torch at runtime.
+
+Extract speaker embeddings, compute cosine similarity, and verify speaker identity
+using ONNX-exported models downloaded automatically from HuggingFace.
+
+**Model collection:** [OpenVoiceOS/speaker-embeddings-onnx](https://huggingface.co/collections/OpenVoiceOS/speaker-embeddings-onnx)
+
+## Install
+
+```bash
+pip install speakeronnx
+```
+
+Optional high-quality resampling:
+```bash
+pip install speakeronnx soxr
+```
+
+## Quick start
+
+```python
+from speakeronnx import SpeakerEmbedder, cosine, verify
+
+embedder = SpeakerEmbedder(model="wespeaker-resnet34")
+
+alice1 = embedder.embed("alice_clip1.wav")
+alice2 = embedder.embed("alice_clip2.wav")
+bob    = embedder.embed("bob_clip1.wav")
+
+print(cosine(alice1, alice2))   # e.g. 0.82  — same speaker
+print(cosine(alice1, bob))      # e.g. 0.21  — different speaker
+
+ok, score = verify(alice1, alice2, threshold=0.45)
+print(ok, score)  # True 0.82
+```
+
+More examples in [`examples/`](examples/).
+
+## CLI
+
+```bash
+speakeronnx list                              # list available models
+speakeronnx embed clip.wav                    # extract embedding
+speakeronnx verify a.wav b.wav               # same-speaker check (exit 0/1)
+speakeronnx verify a.wav b.wav --threshold 0.5
+speakeronnx embed clip.wav --model wespeaker-ecapa512
+```
+
+Full CLI reference in [`docs/cli.md`](docs/cli.md).
+
+## Models
+
+All 9 models are registered in `MODEL_REGISTRY` and downloaded on first use:
+
+| Alias | Embed dim | Frontend | License |
+|---|---|---|---|
+| `wespeaker-resnet34` | 256 | fbank80 | cc-by-4.0 |
+| `wespeaker-ecapa512` | 192 | fbank80 | cc-by-4.0 |
+| `wespeaker-resnet293` | 256 | fbank80 | cc-by-4.0 |
+| `campplus` | 512 | fbank80 | cc-by-4.0 |
+| `campplus-zh-en` | 192 | fbank80 | apache-2.0 |
+| `eres2net` | 192 | fbank80 | apache-2.0 |
+| `titanet-small` | 192 | fbank80 | cc-by-4.0 |
+| `titanet-large` | 192 | fbank80 | cc-by-4.0 |
+| `redimnet-b2` | 192 | raw | apache-2.0 |
+
+Full model comparison and selection guide in [`docs/models.md`](docs/models.md).
+
+## Documentation
+
+| Document | Description |
+|---|---|
+| [`docs/index.md`](docs/index.md) | Full getting-started guide |
+| [`docs/models.md`](docs/models.md) | Model comparison, selection, frontend/layout details |
+| [`docs/api.md`](docs/api.md) | Complete API reference |
+| [`docs/cli.md`](docs/cli.md) | CLI usage reference |
+| [`docs/frontend.md`](docs/frontend.md) | Feature frontend (fbank80 vs raw) technical details |
+| [`docs/advanced.md`](docs/advanced.md) | Custom models, GPU, threshold tuning |
+
+## Examples
+
+| Script | Description |
+|---|---|
+| [`examples/basic_embedding.py`](examples/basic_embedding.py) | Extract embedding from a single WAV |
+| [`examples/verify_speakers.py`](examples/verify_speakers.py) | Verify two clips, try multiple thresholds |
+| [`examples/compare_models.py`](examples/compare_models.py) | Compare all models on same utterances |
+| [`examples/batch_enrollment.py`](examples/batch_enrollment.py) | Enroll speakers from directories, match unknown |
+| [`examples/custom_model.py`](examples/custom_model.py) | Load a custom ONNX model from disk |
+| [`examples/gpu_inference.py`](examples/gpu_inference.py) | CUDA / CoreML inference |
+
+## Tests
+
+```bash
+# Unit tests (mocked, no downloads, no network)
+pytest tests/test_unit.py tests/test_audio.py tests/test_frontend.py \
+      tests/test_embedder.py tests/test_cli.py tests/test_model_registry.py -v
+| Alias | HF repo | License | Embed dim | Description |
+|---|---|---|---|---|
+| `wespeaker-resnet34` | [Wespeaker/wespeaker-voxceleb-resnet34-LM](https://huggingface.co/Wespeaker/wespeaker-voxceleb-resnet34-LM) | cc-by-4.0 | 256 | ResNet34 r-vector, VoxCeleb2 Dev — **recommended default** |
+| `wespeaker-ecapa512` | [Wespeaker/wespeaker-ecapa-tdnn512-LM](https://huggingface.co/Wespeaker/wespeaker-ecapa-tdnn512-LM) | cc-by-4.0 | 192 | ECAPA-TDNN-512 x-vector, VoxCeleb2 Dev |
+| `wespeaker-resnet293` | [Wespeaker/wespeaker-voxceleb-resnet293-LM](https://huggingface.co/Wespeaker/wespeaker-voxceleb-resnet293-LM) | cc-by-4.0 | 256 | ResNet293 r-vector — highest accuracy, 28M params |
+| `campplus` | [csukuangfj/speaker-embedding-models](https://huggingface.co/csukuangfj/speaker-embedding-models) | cc-by-4.0 | 512 | CAM++ (D-TDNN backbone), VoxCeleb2 Dev |
+| `campplus-zh-en` | [csukuangfj/speaker-embedding-models](https://huggingface.co/csukuangfj/speaker-embedding-models) | apache-2.0 | 192 | 3D-Speaker CAM++ multilingual (zh+en) |
+| `eres2net` | [csukuangfj/speaker-embedding-models](https://huggingface.co/csukuangfj/speaker-embedding-models) | apache-2.0 | 192 | ERes2Net, VoxCeleb |
+| `titanet-small` | [csukuangfj/speaker-embedding-models](https://huggingface.co/csukuangfj/speaker-embedding-models) | cc-by-4.0 | 192 | NVIDIA NeMo TitaNet-small (~40 MB) |
+| `titanet-large` | [csukuangfj/speaker-embedding-models](https://huggingface.co/csukuangfj/speaker-embedding-models) | cc-by-4.0 | 192 | NVIDIA NeMo TitaNet-large (~101 MB) |
+| `redimnet-b2` | [OpenVoiceOS/redimnet-b2-vox2-onnx](https://huggingface.co/OpenVoiceOS/redimnet-b2-vox2-onnx) | apache-2.0 | 192 | ReDimNet b2 (1.8M params), raw audio input |
+
+# End-to-end tests (downloads models + generates TTS audio)
+pytest tests/test_e2e.py -v -s
+```
+
+Use `speakeronnx list` to print descriptions and metadata for all registered models.
+
+## Feature frontend
+
+- **fbank80** models: 80-dim log-Mel filterbank with per-utterance CMN,
+  implemented in pure numpy. See [`docs/frontend.md`](docs/frontend.md).
+- **raw** models (redimnet-b2): raw 16 kHz waveform passed directly
+  to ONNX (internal MelSpectrogram in the model).
+
+## Audio requirements
+
+- Mono PCM WAV, any bit depth (8/16/24/32-bit int, 32-bit float)
+- Any sample rate (resampled internally to 16 kHz)
+- Stereo files are downmixed to mono
+- Minimum ~1 second; recommended enrollment 5–30 seconds per speaker
+
+## Dependencies
+
+- `onnxruntime`
+- `numpy`
+- `huggingface_hub`
+- `soxr` (optional, for high-quality resampling)
+
+## Project links
+
+- **GitHub:** [TigreGotico/speakeronnx](https://github.com/TigreGotico/speakeronnx)
+- **PyPI:** `pip install speakeronnx`

speakeronnx-0.0.1/examples/basic_embedding.py

@@ -0,0 +1,30 @@
+"""Basic speaker embedding extraction.
+
+Usage:
+    python examples/basic_embedding.py path/to/speaker.wav
+"""
+import sys
+import numpy as np
+from speakeronnx import SpeakerEmbedder, cosine
+
+
+def main():
+    if len(sys.argv) < 2:
+        print("Usage: python basic_embedding.py <wav_file> [model_alias]")
+        sys.exit(1)
+
+    wav_path = sys.argv[1]
+    model_alias = sys.argv[2] if len(sys.argv) > 2 else "wespeaker-resnet34"
+
+    embedder = SpeakerEmbedder(model=model_alias)
+    embedding = embedder.embed(wav_path)
+
+    print(f"Model : {model_alias}")
+    print(f"Dim   : {len(embedding)}")
+    print(f"Norm  : {np.linalg.norm(embedding):.6f}")
+    print(f"First 8 values: {embedding[:8].tolist()}")
+    print(f"Self-cosine    : {cosine(embedding, embedding):.6f}  (should be 1.0)")
+
+
+if __name__ == "__main__":
+    main()

speakeronnx-0.0.1/examples/batch_enrollment.py

@@ -0,0 +1,84 @@
+"""Batch speaker enrollment and verification.
+
+Enroll multiple speakers from enrollment directories, then verify
+unknown clips against the enrolled gallery.
+
+Directory structure:
+    enrollments/
+        alice/
+            clip1.wav
+            clip2.wav
+        bob/
+            clip1.wav
+        ...
+
+Usage:
+    python examples/batch_enrollment.py enrollments/ test_clip.wav
+"""
+import os
+import sys
+import numpy as np
+from speakeronnx import SpeakerEmbedder, cosine
+
+
+def enroll_speakers(enroll_dir: str, embedder: SpeakerEmbedder):
+    speaker_embs = {}
+
+    for speaker in sorted(os.listdir(enroll_dir)):
+        spk_dir = os.path.join(enroll_dir, speaker)
+        if not os.path.isdir(spk_dir):
+            continue
+
+        embeddings = []
+        for fname in sorted(os.listdir(spk_dir)):
+            if not fname.endswith(".wav"):
+                continue
+            path = os.path.join(spk_dir, fname)
+            emb = embedder.embed(path)
+            embeddings.append(emb)
+            print(f"  Enrolled {speaker}/{fname}  dim={len(emb)}")
+
+        if embeddings:
+            # Average enrollment embeddings (L2-norm the mean)
+            mean_emb = np.mean(embeddings, axis=0)
+            mean_emb /= np.linalg.norm(mean_emb)
+            speaker_embs[speaker] = mean_emb
+            print(f"  → {speaker}: {len(embeddings)} clips, mean emb shape={mean_emb.shape}")
+
+    return speaker_embs
+
+
+def main():
+    if len(sys.argv) < 3:
+        print("Usage: python batch_enrollment.py <enroll_dir/> <test_clip.wav> [model_alias]")
+        sys.exit(1)
+
+    enroll_dir = sys.argv[1]
+    test_wav = sys.argv[2]
+    model_alias = sys.argv[3] if len(sys.argv) > 3 else "wespeaker-resnet34"
+
+    embedder = SpeakerEmbedder(model=model_alias)
+
+    print(f"Enrolling speakers from {enroll_dir}/ ...")
+    gallery = enroll_speakers(enroll_dir, embedder)
+
+    print(f"\nVerifying {test_wav} against gallery ...")
+    test_emb = embedder.embed(test_wav)
+
+    results = []
+    for speaker, enroll_emb in gallery.items():
+        score = cosine(test_emb, enroll_emb)
+        results.append((score, speaker))
+
+    results.sort(key=lambda x: -x[0])
+    print(f"\n{'Speaker':<20} {'Score':>10}")
+    print("-" * 32)
+    for score, speaker in results:
+        print(f"{speaker:<20} {score:>10.6f}")
+
+    best = results[0]
+    print(f"\nBest match: {best[1]} (score={best[0]:.4f})")
+
+
+if __name__ == "__main__":
+    main()

speakeronnx-0.0.1/examples/compare_models.py

@@ -0,0 +1,52 @@
+"""Compare multiple models on the same audio files.
+
+Shows how different models perform on the same pair of utterances.
+Useful for selecting the right model for your data.
+
+Usage:
+    python examples/compare_models.py <same1.wav> <same2.wav> <diff1.wav>
+"""
+import sys
+import numpy as np
+from speakeronnx import SpeakerEmbedder, cosine, MODEL_REGISTRY
+
+# Skip models that take raw audio (need different handling)
+SKIP_RAW = True
+
+
+def main():
+    if len(sys.argv) < 4:
+        print("Usage: python compare_models.py <same1.wav> <same2.wav> <diff1.wav>")
+        sys.exit(1)
+
+    same1, same2, diff = sys.argv[1], sys.argv[2], sys.argv[3]
+
+    print(f"{'Model':<22} {'Same-spk':>10} {'Diff-spk':>10} {'Margin':>10}  {'Dim':>4}")
+    print("-" * 60)
+
+    results = []
+    for alias, entry in MODEL_REGISTRY.items():
+        if SKIP_RAW and entry.frontend == "raw":
+            continue
+
+        emb = SpeakerEmbedder(model=alias)
+
+        e_same1 = emb.embed(same1)
+        e_same2 = emb.embed(same2)
+        e_diff = emb.embed(diff)
+
+        same_score = cosine(e_same1, e_same2)
+        diff_score = cosine(e_same1, e_diff)
+        margin = same_score - diff_score
+
+        results.append((alias, same_score, diff_score, margin, entry.embed_dim))
+
+    results.sort(key=lambda r: -r[3])  # sort by margin descending
+    for alias, same, diff, margin, dim in results:
+        print(f"{alias:<22} {same:>10.4f} {diff:>10.4f} {margin:>10.4f}  {dim:>4}")
+
+    print("\nModels sorted by same-vs-different margin (higher = better separation)")
+
+
+if __name__ == "__main__":
+    main()

speakeronnx-0.0.1/examples/custom_model.py

@@ -0,0 +1,36 @@
+"""Load a custom ONNX model from disk.
+
+Usage:
+    python examples/custom_model.py /path/to/model.onnx clip.wav
+"""
+import sys
+import numpy as np
+from speakeronnx import SpeakerEmbedder, cosine
+
+
+def main():
+    if len(sys.argv) < 3:
+        print("Usage: python custom_model.py <model.onnx> <wav_file> [wav_file2]")
+        sys.exit(1)
+
+    onnx_path = sys.argv[1]
+    wav1_path = sys.argv[2]
+
+    embedder = SpeakerEmbedder(model=onnx_path)
+    print(f"Model        : {onnx_path}")
+    print(f"Sample rate  : {embedder.sample_rate} Hz")
+    print(f"Embed dim    : {embedder.embed_dim}")
+    print(f"Is registered: {embedder.entry is not None}")
+
+    emb1 = embedder.embed(wav1_path)
+    print(f"Embedding    : dim={len(emb1)}  norm={np.linalg.norm(emb1):.6f}")
+
+    if len(sys.argv) >= 4:
+        wav2_path = sys.argv[3]
+        emb2 = embedder.embed(wav2_path)
+        score = cosine(emb1, emb2)
+        print(f"Cosine({wav1_path}, {wav2_path}): {score:.6f}")
+
+
+if __name__ == "__main__":
+    main()

speakeronnx-0.0.1/examples/gpu_inference.py

@@ -0,0 +1,37 @@
+"""GPU inference example (requires onnxruntime-gpu).
+
+Usage:
+    python examples/gpu_inference.py clip.wav
+"""
+import sys
+import numpy as np
+from speakeronnx import SpeakerEmbedder
+
+
+def main():
+    if len(sys.argv) < 2:
+        print("Usage: python gpu_inference.py <wav_file>")
+        sys.exit(1)
+
+    wav_path = sys.argv[1]
+
+    # Try CUDA, fall back to CPU
+    providers = [
+        "CUDAExecutionProvider",
+        "CoreMLExecutionProvider",
+        "CPUExecutionProvider",
+    ]
+
+    embedder = SpeakerEmbedder(
+        model="wespeaker-resnet34",
+        providers=providers,
+    )
+
+    embedding = embedder.embed(wav_path)
+    print(f"Embedding dim : {len(embedding)}")
+    print(f"L2 norm       : {np.linalg.norm(embedding):.6f}")
+    print(f"Provider used : {embedder._session.get_providers()}")
+
+
+if __name__ == "__main__":
+    main()

speakeronnx-0.0.1/examples/verify_speakers.py

@@ -0,0 +1,47 @@
+"""Speaker verification with threshold tuning.
+
+Demonstrates how to verify whether two audio files are from the same
+speaker and tune the decision threshold.
+
+Usage:
+    python examples/verify_speakers.py <enroll.wav> <test.wav>
+"""
+import sys
+import numpy as np
+from speakeronnx import SpeakerEmbedder, cosine, verify
+
+
+def main():
+    if len(sys.argv) < 3:
+        print("Usage: python verify_speakers.py <enroll.wav> <test.wav> [model_alias]")
+        sys.exit(1)
+
+    enroll_wav = sys.argv[1]
+    test_wav = sys.argv[2]
+    model_alias = sys.argv[3] if len(sys.argv) > 3 else "wespeaker-resnet34"
+
+    embedder = SpeakerEmbedder(model=model_alias)
+
+    enroll_emb = embedder.embed(enroll_wav)
+    test_emb = embedder.embed(test_wav)
+
+    score = cosine(enroll_emb, test_emb)
+    print(f"Model  : {model_alias}")
+    print(f"Score  : {score:.6f}")
+
+    # Try multiple thresholds
+    for threshold in [0.3, 0.4, 0.45, 0.5, 0.6]:
+        ok, _ = verify(enroll_emb, test_emb, threshold=threshold)
+        print(f"  threshold={threshold:.2f} → {'SAME' if ok else 'DIFFERENT'}")
+
+    # Recommendation
+    if score > 0.5:
+        print("\nLikely SAME speaker (score > 0.5)")
+    elif score < 0.3:
+        print("\nLikely DIFFERENT speaker (score < 0.3)")
+    else:
+        print("\nAmbiguous — tune threshold on held-out data")
+
+
+if __name__ == "__main__":
+    main()

speakeronnx-0.0.1/pyproject.toml

@@ -0,0 +1,53 @@
+[build-system]
+requires = ["setuptools>=68", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "speakeronnx"
+dynamic = ["version"]
+description = "Pure-onnxruntime speaker embedding library — no torch at runtime"
+license = {text = "Apache-2.0"}
+classifiers = [
+    "Programming Language :: Python :: 3",
+    "License :: OSI Approved :: Apache Software License",
+    "Operating System :: OS Independent",
+]
+authors = [{name = "JarbasAI", email = "jarbasai@mailfence.com"}]
+readme = "README.md"
+requires-python = ">=3.9"
+dependencies = [
+    "numpy",
+    "onnxruntime",
+    "huggingface_hub",
+]
+
+[project.optional-dependencies]
+soxr = ["soxr"]
+test = [
+    "pytest",
+    "edge-tts",
+    "requests",
+]
+
+[project.scripts]
+speakeronnx = "speakeronnx.__main__:main"
+
+[project.urls]
+Homepage = "https://github.com/TigreGotico/speakeronnx"
+
+[tool.setuptools.dynamic]
+version = {attr = "speakeronnx.version.__version__"}
+
+[tool.setuptools.packages.find]
+where = ["."]
+
+[tool.setuptools.package-data]
+"speakeronnx" = ["**/*"]
+
+[tool.mypy]
+python_version = "3.9"
+ignore_missing_imports = true
+no_site_packages = true
+strict = false
+warn_unused_ignores = true
+warn_return_any = false

speakeronnx-0.0.1/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+