SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Revert "Incorrect Whisper long-form decoding timestamps " (#32148)

Browse Source 
Revert "Incorrect Whisper long-form decoding timestamps  (#32003)"

This reverts commit cd48553fc8.
This commit is contained in:
Sanchit Gandhi
2024-07-23 18:34:30 +08:00
committed by GitHub
parent 034b477847
commit 3263b34354
5 changed files with 25 additions and 133 deletions
Download Patch File Download Diff File Expand all files Collapse all files

66
tests/models/whisper/test_modeling_whisper.py
View File

@@ -2243,72 +2243,6 @@ class WhisperModelIntegrationTests(unittest.TestCase):
transcript = processor.batch_decode(generated_ids, skip_special_tokens=True, output_offsets=True)
self.assertEqual(transcript, EXPECTED_TRANSCRIPT)
@slow
def test_tiny_longform_timestamps_generation(self):
set_seed(0)
processor = WhisperProcessor.from_pretrained("openai/whisper-tiny")
model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny")
model.to(torch_device)
sample = self._load_datasamples(1)
input_speech = np.concatenate(sample * 10)
input_features = processor(input_speech, return_tensors="pt", truncation=False, sampling_rate=16_000)
input_features = input_features.to(torch_device)
generated_ids = model.generate(**input_features, return_timestamps=True, return_segments=True)
EXPECTED_TRANSCRIPT = [
{
"text": " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel. Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel. Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel. Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel. Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel. Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel. Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel. Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel. Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel. Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.",
"offsets": [
{
"text": " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.",
"timestamp": (0.0, 6.0),
},
{
"text": " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.",
"timestamp": (6.0, 12.0),
},
{
"text": " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.",
"timestamp": (12.0, 18.0),
},
{
"text": " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.",
"timestamp": (18.0, 24.0),
},
{
"text": " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.",
"timestamp": (24.0, 29.0),
},
{
"text": " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.",
"timestamp": (29.0, 35.0),
},
{
"text": " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.",
"timestamp": (35.0, 41.0),
},
{
"text": " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.",
"timestamp": (41.0, 47.0),
},
{
"text": " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.",
"timestamp": (47.0, 53.0),
},
{
"text": " Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.",
"timestamp": (53.0, 58.20000076293945),
},
],
}
]
transcript = processor.batch_decode(generated_ids, skip_special_tokens=True, output_offsets=True)
self.assertEqual(transcript, EXPECTED_TRANSCRIPT)
@slow
def test_large_timestamp_generation(self):
set_seed(0)