INNER CODE UNIT · Python

chunk_end_idx

sanchit-gandhi/whisper-jax · benchmarks/run_pipeline_dataloader.py:25

            chunk_end_idx = chunk_start_idx + chunk_len
            chunk = inputs[chunk_start_idx:chunk_end_idx]
            processed = feature_extractor(chunk, sampling_rate=feature_extractor.sampling_rate, return_tensors="np")
            _stride_left = 0 if chunk_start_idx == 0 else stride_left
            # all right strides must be full, otherwise it is the last item
            is_last = chunk_end_idx > inputs_len if stride_right > 0 else chunk_end_idx >= inputs_len
            _stride_right = 0 if is_last else stride_right

            chunk_len = chunk.shape[0]
            stride = (chunk_len, _stride_left, _stride_right)
            if chunk.shape[0] > _stride_left:
                yield {"is_last": is_last, "stride": stride, **processed}
            if is_last:
                break

    def preprocess(self, inputs, chunk_length_s=0, stride_length_s=None):
        array = inputs.get("array")
        in_sampling_rate = inputs.get("sampling_rate")

View source record →

📰 Research Paper
Loading…
⏳ Fetching content…