univ_conformer_small_ge2e_contrast

FlaxSpeaker: univ_conformer_small_ge2e_contrast (conformer - small)

Disclaimer & Reproduction Notice

This is not an officially supported Google product.

This checkpoint and the accompanying FlaxSpeaker library are a from-scratch re-implementation in JAX/Flax, written using only the published papers (no proprietary code or data), of speaker recognition algorithms from: - 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) - ATP (pooling head repurposed from a language identification paper): 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:2203.05642)

The experiments in those papers were run on Google-internal infrastructure and data, not with this library. This is a 400-step demonstration checkpoint, not a reproduction of the papers' results -- see Training budget and Deviations from the source papers below.

Training Budget (read this first)

This model was trained for 400 gradient steps with mini-batches of 32 utterances (8 speakers x 4 utterances), i.e. about 1.35 epochs over the 9,447-utterance universal corpus, on a CPU workstation, with a single seed (42). Losses are not converged at this budget. All EERs below are therefore far from state of the art and are only meaningful for relative comparisons between FlaxSpeaker checkpoints trained under the same budget.

Training-crop provenance. This checkpoint was trained with U(90, 180)-frame variable-length crops and an exact validity mask (the current pipeline default); the shipped config.json / YAML (variable_length_training: true) describes the released weights. Step-0 loss fingerprint at the released seed (scripts/fingerprint_released_checkpoints.py, pretrained_models/step0_fingerprints.json): archived 1.0001; current code with fixed 180-frame crops 1.0001, with U(90, 180) crops 1.0001 (match: both). Both crop regimes reproduce the archived 4-decimal step-0 loss (fixed 1.0001, variable 1.0001); the crop regime is assigned from the git history, not from the fingerprint. Evaluating the released weights with the current code reproduces the archived metrics in eval_results.json. The seed-to-seed standard deviation of pooled 1-utt EER at this budget is about 0.7-0.8 % absolute, so sub-1.5 % differences between checkpoints are not rankings.

For orientation only: published ECAPA-TDNN systems trained on VoxCeleb2-dev reach about 0.9% EER on the full VoxCeleb1-O list (Desplanques et al., 2020), and sub-1M-parameter systems about 2% (Wang et al., EfficientTDNN, 2022). The 400-step checkpoints released here are 10-30x worse than that and are intended for relative comparisons within the FlaxSpeaker benchmark, not as state-of-the-art models.

Model Summary

Model Files Included

• model.safetensors: model weights (encoder + GE2E w,b / Dr-Vector / PFAS heads) in SafeTensors format
• flax_model.msgpack: native Flax Msgpack serialized parameters
• config.json: FlaxSpeakerConfig (full ExperimentConfig, incl. the frontend/VAD settings)
• eval_results.json: multi-domain evaluation metrics (1-utt & 3-utt EER, minDCF, ROC-AUC, 95% bootstrap CIs, VoxCeleb1-O 10-speaker subset)

Quickstart Usage

Python API (from Hugging Face Hub or Local Directory)

Requires flaxspeaker >= 0.3.0 (pip install -U flaxspeaker); the 0.2.0 wheel on PyPI predates the released checkpoints and cannot load them.

from flaxspeaker import load_pretrained

model = load_pretrained("wq2012/univ_conformer_small_ge2e_contrast")
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/univ_conformer_small_ge2e_contrast \
    --enroll_audio path/to/enroll.flac --test_audio path/to/test.flac

Multi-Domain Evaluation Results

1-Utterance Verification Across Evaluation Domains

Trial lists are 50/50 target/impostor over held-out speakers (LibriSpeech 40, CN-Celeb2 30, VoxCeleb1 10 speakers); [email protected] is close to its trivial ceiling of 1.0 for every model under this trial mix.

Caveat: VoxCeleb1-O 10-speaker subset: the official veri_test2.txt list filtered to trials whose enrollment and test speakers are both among the 10 held-out evaluation speakers (2,500 trials, ~72% target). Not comparable to published VoxCeleb1-O numbers; the other 30 of the 40 official test speakers are used as training data by vox_* and univ_* models.

3-Utterance Multi-Enrollment Verification Across Evaluation Domains

Single seed; the 95% bootstrap CIs resample trials (not speakers). Differences below about 1% absolute EER between FlaxSpeaker checkpoints are within the seed-to-seed variation measured in pretrained_models/r2_seed_and_control_results.json (see the GitHub README) and should not be interpreted as rankings.

Deviations from the Source Papers

• Training regime: 400 gradient steps x 32 utterances (N = 8 speakers x M = 4) on <= 25 utterances/speaker from 381 (universal) or fewer speakers; the source papers train for ~500k steps with N = 16-64 / M = 8-10 on thousands of speakers.
• Augmentation: synthetic log-Mel-domain noise + one-tap smoothing (specaug.apply_feature_mtr_augmentation), not waveform MTR with recorded noise/RIR corpora.
• Objective: hybrid contrast surrogate (gradient = 0.25 Eq. 7 + 0.25 BCE + 0.50 softmax-CE) with ge2e_init_b = -1, not Wan et al. Eq. 7 alone. Set ge2e_contrast_surrogate: false for the pure Eq. 7 gradient.
• Conformer: parallel linear + Conv1D stem, no positional encoding, 3 blocks x 192-D, 4 heads, kernel 15, mask-aware GroupNorm(2), no dropout, multi-layer feature aggregation; the Google papers use 12 x 512 blocks, 4-8 heads, kernel 32, relative positional encoding and dropout.
• ATP pooling: 4 heads with a tanh bottleneck (128-D) and subspace_heads = True; the papers use a single head with a linear-sigmoid weight (attention_dim = 0). Note that arXiv:2202.12163 is a language-identification paper whose pooling head is repurposed here.
• Frontend: 80-D natural-log Mel + frame-level energy VAD (26 dB dynamic range) + CMVN; evaluation centre-crops to at most 500 frames (5 s).

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}
}