h-j-han/SpeechQE-CoVoST2

SpeechQE: Estimating the Quality of Direct Speech Translation

This is a benchmark and training corpus for the task of quality estimation for speech translation (SpeechQE).

We subsample about 80k segments from the training set and 500 from the dev and test of CoVoST2, then run seven different direct ST models to generate the ST hypotheses. So,test split consists of 3500 instances(500*7). We also provide splits for each translation model. *(We provide test split first, and the training corpus will be provided later. However, if you want those quickly, please do not hesitate to ping me (hjhan@umd.edu)!)

E2E Model Trained with SpeechQE-CoVoST2

Task	E2E Model	Trained Domain
SpeechQE for English-to-German Speech Translation	h-j-han/SpeechQE-TowerInstruct-7B-en2de	CoVoST2
SpeechQE for Spanish-to-English Speech Translation	h-j-han/SpeechQE-TowerInstruct-7B-es2en	CoVoST2

Setup

We provide code in Github repo : https://github.com/h-j-han/SpeechQE

$ git clone https://github.com/h-j-han/SpeechQE.git
$ cd SpeechQE

$ conda create -n speechqe Python=3.11 pytorch=2.0.1  pytorch-cuda=11.7 torchvision torchaudio -c pytorch -c nvidia
$ conda activate speechqe
$ pip install -r requirements.txt

Download Audio Data

Download the audio data from Common Voice. Here, we use mozilla-foundation/common_voice_4_0.

import datasets
cv4en = datasets.load_dataset(
    "mozilla-foundation/common_voice_4_0", "es", cache_dir='path/to/cv4/download',
)

Evaluation with SpeechQE-CoVoST2

We provide SpeechQE benchmark: h-j-han/SpeechQE-CoVoST2. BASE_AUDIO_PATH is the path of downloaded Common Voice dataset.

$ python speechqe/score_speechqe.py \
    --speechqe_model=h-j-han/SpeechQE-TowerInstruct-7B-es2en \
    --dataset_name=h-j-han/SpeechQE-CoVoST2 \
    --base_audio_path=$BASE_AUDIO_PATH \
    --dataset_config_name=es2en \
    --test_split_name=test \

Reference

Please find details in this EMNLP24 paper :

@misc{han2024speechqe,
    title={SpeechQE: Estimating the Quality of Direct Speech Translation},
    author={HyoJung Han and Kevin Duh and Marine Carpuat},
    year={2024},
    eprint={2410.21485},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}