facebook/wav2vec2-xls-r-300m 在 Igbo 语言上进行了对 google/fleurs 和 mozilla-foundation/common_voice_13_0 的 fine-tuned。

WER: 0.51

运行代码:

from huggingsound import SpeechRecognitionModel

model = SpeechRecognitionModel("AstralZander/igbo_ASR")
audio_paths = [audio_path] # List with paths to audio
transcriptions = model.transcribe(audio_paths)

transcriptions # List of transcriptions, timestamps and probabilities
transcriptions[ind_audio]['transcription'] # Transcription of audio with the ind_audio index from the audio_paths list