Reza2kn/Shenava-Rizeh-v1.0-CoreML-iOS15-fp16

1{2  "schema": "visualears-preprocessor-v2-nemo-fbank",3  "description": "NeMo AudioToMelSpectrogramPreprocessor-compatible browser feature pipeline for shenava_koochik_1_0. normalize=NA, preemphasis=0.97, torch.stft center=True, Slaney mel filters, requires processed_signal_length.",4  "sample_rate": 16000,5  "n_fft": 512,6  "win_length": 400,7  "hop_length": 160,8  "n_mels": 80,9  "window": "hann_periodic_false",10  "window_formula": "0.5 - 0.5*cos(2*pi*i/(win_length-1))",11  "center": true,12  "center_pad": 256,13  "pad_mode": "reflect",14  "preemphasis": 0.97,15  "mel_scale": "slaney/librosa.filters.mel(htk=False,norm='slaney')",16  "mel_filters_file": "mel_filters_slaney_80x257.json",17  "spectrum": "magnitude_power_2_no_fft_normalization",18  "log": "natural",19  "log_zero_guard_type": "add",20  "log_zero_guard_value": 5.960464477539063e-08,21  "normalize": "NA",22  "fixed_frames": 2005,23  "pad_value": 0.0,24  "frame_count_formula": "max(1, min(fixed_frames, floor(num_samples / hop_length) + 1))",25  "inputs": [26    {27      "name": "processed_signal",28      "dtype": "float16",29      "shape": [30        "batch",31        80,32        200533      ]34    },35    {36      "name": "processed_signal_length",37      "dtype": "int64",38      "shape": [39        "batch"40      ]41    }42  ],43  "outputs": [44    {45      "name": "logits",46      "dtype": "float16",47      "shape": [48        "batch",49        252,50        102551      ]52    },53    {54      "name": "encoded_lengths",55      "dtype": "int64",56      "shape": [57        "batch"58      ]59    }60  ],61  "output_stride": 8,62  "usable_steps_formula": "min(encoded_lengths[0], logits_steps)",63  "ms_per_output_step": 80,64  "blank_id": 1024,65  "ctc_decode": "greedy argmax; drop blank, repeats, and SentencePiece special tokens; join pieces; '▁' -> space"66}67