FlaxSpeaker: `ls_conformer_small_dr_vectors` (`conformer` - `small`)

Disclaimer & Open-Source Reproduction Notice

This is not an officially supported Google product.

This model checkpoint and the accompanying FlaxSpeaker library are an independent, clean-room open-source reproduction in JAX/Flax of speaker recognition algorithms published in:

GE2E: Generalized End-to-End Loss for Speaker Verification (arXiv:1710.10467)

Dr-Vectors: Decision Residual Networks and an Improved Loss for Speaker Recognition (arXiv:2104.01989)

Conformer-ATP: Attentive Temporal Pooling for Conformer-based Streaming Language Identification in Long-form Speech (arXiv:2202.12163)

PFAS: Parameter-Free Attentive Scoring for Speaker Verification (arXiv:2203.05642v3)

The experiments in those original papers were conducted using Google's proprietary internal infrastructure and internal datasets, not this open-source library or public dataset split.

Model Summary

Property	Value
Model ID	`ls_conformer_small_dr_vectors`
Hugging Face Repository	`wq2012/ls_conformer_small_dr_vectors`
Hugging Face Collection	FlaxSpeaker Collection
GitHub Repository	`wq2012/FlaxSpeaker`
PyPI Package	`flaxspeaker`
Backbone Architecture	`conformer` (`small`)
Temporal Pooling	`atp`
Training Loss	`extended_set_softmax`
Verification Scoring	`dr_vector`
Output Embedding Dim	`192`
Encoder Parameters	`2,088,007`
Training Domain	`librispeech`
Audio Frontend	`80`-dim `log_mel` @ `16000` Hz + Energy VAD + CMVN

Model Files Included

model.safetensors: Primary model weights in SafeTensors format
flax_model.msgpack: Native Flax Msgpack serialized parameters
config.json: FlaxSpeakerConfig model & training architecture configuration
preprocessor_config.json: WhisperFeatureExtractor-compatible frontend configuration
eval_results.json: Verified multi-domain evaluation metrics (1-utt & 3-utt EER, minDCF, ROC-AUC, 95% bootstrap CIs)

Quickstart Usage

Python API (from Hugging Face Hub or Local Directory)

from flaxspeaker import load_pretrained

model = load_pretrained("wq2012/ls_conformer_small_dr_vectors")
emb = model.extract_embedding("path/to/utterance.flac")
score = model.verify("path/to/enroll.flac", "path/to/test.flac")
print(f"Embedding shape: {emb.shape}, Verification score: {score:.4f}")

Command-Line Inference

python scripts/inference.py --model_dir wq2012/ls_conformer_small_dr_vectors \
    --enroll_audio path/to/enroll.flac --test_audio path/to/test.flac

Multi-Domain Evaluation Results

1-Utterance Verification Across Evaluation Domains

Evaluation Split	Trials	EER (%)	95% Bootstrap CI (%)	minDCF@0.01	minDCF@0.10	ROC-AUC
LibriSpeech (English Read)	3,000	11.17%	[9.87, 12.20]	0.9187	0.5340	0.9565
CN-Celeb2 (Chinese Multi-Genre)	3,000	28.80%	[27.28, 30.40]	0.9960	0.9207	0.7933
VoxCeleb1 (In-the-Wild Celebrity)	2,000	22.30%	[20.72, 24.43]	0.9850	0.9370	0.8524
Universal Pooled (All 3 Domains)	8,000	20.69%	[19.82, 21.60]	0.9925	0.8147	0.8784
Universal Macro-Average (3 Domains)	8,000	20.76%	—	0.9666	0.7972	0.8674

3-Utterance Multi-Enrollment Verification Across Evaluation Domains

Evaluation Split	Trials	3-Utt EER (%)	95% Bootstrap CI (%)	minDCF@0.01	ROC-AUC
LibriSpeech (3-Utt)	1,000	6.20%	[4.80, 7.80]	0.3980	0.9851
CN-Celeb2 (3-Utt)	1,000	24.00%	[21.40, 26.15]	0.9240	0.8445
VoxCeleb1 (3-Utt)	1,000	19.60%	[17.05, 21.80]	0.9780	0.8880
Universal Pooled (3-Utt)	3,000	16.60%	[15.33, 17.82]	0.9607	0.9157
Universal Macro-Average (3-Utt)	3,000	16.60%	—	0.7667	0.9059

Citation

@inproceedings{wan2018generalized,
  title={Generalized End-to-End Loss for Speaker Verification},
  author={Wan, Li and Wang, Quan and Papir, Alan and Moreno, Ignacio Lopez},
  booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={4879--4883},
  year={2018}
}

@inproceedings{pelecanos2021dr,
  title={Dr-Vectors: Decision Residual Networks and an Improved Loss for Speaker Recognition},
  author={Pelecanos, Jason and Wang, Quan and Moreno, Ignacio Lopez},
  booktitle={Interspeech},
  pages={4603--4607},
  year={2021}
}

@inproceedings{wang2022attentive,
  title={Attentive Temporal Pooling for Conformer-based Streaming Language Identification in Long-form Speech},
  author={Wang, Quan and Yu, Yang and Pelecanos, Jason and Huang, Yiling and Moreno, Ignacio Lopez},
  booktitle={The Speaker and Language Recognition Workshop (Odyssey)},
  year={2022}
}

@inproceedings{pelecanos2022parameter,
  title={Parameter-Free Attentive Scoring for Speaker Verification},
  author={Pelecanos, Jason and Wang, Quan and Huang, Yiling and Moreno, Ignacio Lopez},
  booktitle={The Speaker and Language Recognition Workshop (Odyssey)},
  year={2022}
}