not using torchaudio for reading anymore

CHANGELOG.md

@@ -7,7 +7,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
 
 ## [1.0.1] - TBD
 
-Not using torchaudio anymore when writing audio files, relying instead directly on the commandline ffmpeg.
+Not using torchaudio anymore when writing audio files, relying instead directly on the commandline ffmpeg. Also not using it anymore for reading audio files, for similar reasons.
 
 ## [1.0.0] - 2023-09-07
 

audiocraft/data/audio.py

@@ -137,12 +137,6 @@ def audio_read(filepath: tp.Union[str, Path], seek_time: float = 0.,
         wav = torch.from_numpy(wav).t().contiguous()
         if len(wav.shape) == 1:
             wav = torch.unsqueeze(wav, 0)
-    elif (
-        fp.suffix in ['.wav', '.mp3'] and fp.suffix[1:] in ta.utils.sox_utils.list_read_formats()
-        and duration <= 0 and seek_time == 0
-    ):
-        # Torchaudio is faster if we load an entire file at once.
-        wav, sr = ta.load(fp)
     else:
         wav, sr = _av_read(filepath, seek_time, duration)
     if pad and duration > 0: