tflite-hub/conformer-lang-id

Conformer based spoken language identification model

Summary

This is a conformer-based streaming language identification model with attentive temporal pooling.

The model was trained with public data only.

The paper: https://arxiv.org/abs/2202.12163

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

Usage

Run use this model, you will need to use the siglingvo library: https://github.com/google/speaker-id/tree/master/lingvo

Since lingvo does not support Python 3.11 yet, make sure your Python is up to 3.10.

Install the library:

pip install sidlingvo

Example usage:

import os
from sidlingvo import wav_to_lang
from huggingface_hub import hf_hub_download

repo_id = "tflite-hub/conformer-lang-id"
model_path = "models"
hf_hub_download(repo_id=repo_id, filename="vad_short_model.tflite", local_dir=model_path)
hf_hub_download(repo_id=repo_id, filename="vad_short_mean_stddev.csv", local_dir=model_path)
hf_hub_download(repo_id=repo_id, filename="conformer_langid_medium.tflite", local_dir=model_path)

wav_file = "your_wav_file.wav"
runner = wav_to_lang.WavToLangRunner(
    vad_model_file=os.path.join(model_path, "vad_short_model.tflite"),
    vad_mean_stddev_file=os.path.join(model_path, "vad_short_mean_stddev.csv"),
    langid_model_file=os.path.join(model_path, "conformer_langid_medium.tflite"))
top_lang, _ = runner.wav_to_lang(wav_file)
print("Predicted language:", top_lang)