[w2vbert] support w2vbert fbank (wenet-e2e#2346)

* [ssl/w2vbert] support w2vbert fbank

* [ssl/w2vbert] add libsndfile in ut yaml

.github/workflows/unit_test.yml

@@ -40,7 +40,7 @@ jobs:
         run: |
           set -eux
           pip install -r requirements.txt
-          sudo apt update && sudo apt install -y ffmpeg  libsox-dev
+          sudo apt update && sudo apt install -y ffmpeg  libsox-dev libsndfile1
       - name: Run Pytest
         run: |
           set -eux

test/wenet/ssl/w2vbert/test_w2vbert.py

@@ -0,0 +1,50 @@
+from pathlib import Path
+import pytest
+import torch
+import torchaudio
+
+from wenet.dataset import processor
+
+try:
+    import fairseq2  # noqa
+    from fairseq2.data.audio import AudioDecoder, WaveformToFbankConverter
+    from fairseq2.memory import MemoryBlock
+except ImportError:
+    import os
+    os.system('pip install --no-input fairseq2')
+    import fairseq2  # noqa
+    from fairseq2.data.audio import AudioDecoder, WaveformToFbankConverter
+    from fairseq2.memory import MemoryBlock
+
+
+@pytest.mark.parametrize(
+    "wav_file",
+    [
+        # "test/resources/aishell-BAC009S0724W0121.wav",
+        "test/resources/librispeech-1995-1837-0001.wav",
+    ])
+def test_w2vbert_fbank(wav_file):
+    fbank_convert = WaveformToFbankConverter(
+        num_mel_bins=80,
+        waveform_scale=2**15,
+        channel_last=True,
+        standardize=True,
+    )
+    audio_decoder = AudioDecoder(dtype=torch.float32)
+    with Path(wav_file).open("rb") as fb:
+        block = MemoryBlock(fb.read())
+    decode_audio = audio_decoder(block)
+    w2vbert_waveform = decode_audio['waveform']
+    w2vbert_mat = fbank_convert(decode_audio)['fbank']
+
+    wenet_waveform, _ = torchaudio.load(wav_file)
+    fbank_args = {
+        "num_mel_bins": 80,
+        "frame_length": 25,
+        "frame_shift": 10,
+        "dither": 0.0,
+    }
+    sample = {'sample_rate': 16000, "wav": wenet_waveform, 'key': wav_file}
+    wenet_mat = processor.compute_w2vbert_fbank(sample, **fbank_args)['feat']
+    assert torch.allclose(w2vbert_waveform.transpose(0, 1), wenet_waveform)
+    assert torch.allclose(w2vbert_mat, wenet_mat, atol=9e-5, rtol=9e-4)

wenet/dataset/processor.py

@@ -222,6 +222,22 @@ def compute_fbank(sample,
     return sample
 
 
+def compute_w2vbert_fbank(sample,
+                          num_mel_bins=23,
+                          frame_length=25,
+                          frame_shift=10,
+                          dither=0.0):
+    """ Extract Pretrain w2vbert(4.5M hours) fbank
+    """
+    sample = compute_fbank(sample, num_mel_bins, frame_length, frame_shift,
+                           dither)
+    mat = sample['feat']
+    std, mean = torch.std_mean(mat, dim=0)
+    mat = mat.subtract(mean).divide(std)
+    sample['feat'] = mat
+    return sample
+
+
 def sort_by_feats(sample):
     assert 'feat' in sample
     assert isinstance(sample['feat'], torch.Tensor)