| { |
| "schema": "visualears-preprocessor-v2-nemo-fbank", |
| "description": "NeMo AudioToMelSpectrogramPreprocessor-compatible browser feature pipeline for shenava_rizeh_0_9. normalize=NA, preemphasis=0.97, torch.stft center=True, Slaney mel filters, requires processed_signal_length.", |
| "sample_rate": 16000, |
| "n_fft": 512, |
| "win_length": 400, |
| "hop_length": 160, |
| "n_mels": 80, |
| "window": "hann_periodic_false", |
| "window_formula": "0.5 - 0.5*cos(2*pi*i/(win_length-1))", |
| "center": true, |
| "center_pad": 256, |
| "pad_mode": "reflect", |
| "preemphasis": 0.97, |
| "mel_scale": "slaney/librosa.filters.mel(htk=False,norm='slaney')", |
| "mel_filters_file": "mel_filters_slaney_80x257.json", |
| "spectrum": "magnitude_power_2_no_fft_normalization", |
| "log": "natural", |
| "log_zero_guard_type": "add", |
| "log_zero_guard_value": 5.960464477539063e-08, |
| "normalize": "NA", |
| "fixed_frames": 2005, |
| "pad_value": 0.0, |
| "frame_count_formula": "max(1, min(fixed_frames, floor(num_samples / hop_length) + 1))", |
| "inputs": [ |
| { |
| "name": "processed_signal", |
| "dtype": "float16", |
| "shape": [ |
| "batch", |
| 80, |
| 2005 |
| ] |
| }, |
| { |
| "name": "processed_signal_length", |
| "dtype": "int64", |
| "shape": [ |
| "batch" |
| ] |
| } |
| ], |
| "outputs": [ |
| { |
| "name": "logits", |
| "dtype": "float16", |
| "shape": [ |
| "batch", |
| 252, |
| 1025 |
| ] |
| }, |
| { |
| "name": "encoded_lengths", |
| "dtype": "int64", |
| "shape": [ |
| "batch" |
| ] |
| } |
| ], |
| "output_stride": 8, |
| "usable_steps_formula": "min(encoded_lengths[0], logits_steps)", |
| "ms_per_output_step": 80, |
| "blank_id": 1024, |
| "ctc_decode": "greedy argmax; drop blank, repeats, and SentencePiece special tokens; join pieces; '▁' -> space" |
| } |
|
|
