Add Pocket (300M, CPU) variant: int8/fp32 ONNX, PyTorch checkpoint, vocab, CPU inference script, results; README variants table
19eb87a verified | { | |
| "name": "dynamic-asr", | |
| "version": "0.1.0", | |
| "pipeline_tag": "automatic-speech-recognition", | |
| "library_name": "fairseq2 / omnilingual-asr", | |
| "base_model": "facebook omniASR_CTC_1B_v2 (Apache-2.0)", | |
| "architecture": "wav2vec2 CTC, 1B (unchanged; vocab 10288, tokenizer omniASR_tokenizer_written_v2)", | |
| "languages": [ | |
| "en", | |
| "pcm", | |
| "ig", | |
| "yo", | |
| "ha" | |
| ], | |
| "default_model": "real", | |
| "models": { | |
| "real": { | |
| "path": "real/model.pt", | |
| "trained_on": "licensed real speech only", | |
| "status": "default" | |
| }, | |
| "codeswitch-experimental": { | |
| "path": "codeswitch-experimental/model.pt", | |
| "trained_on": "same real speech + synthetic code-switched audio (Flash-DPO TTS)", | |
| "status": "experimental" | |
| }, | |
| "pocket": { | |
| "path": "pocket/model_int8.onnx", | |
| "alt_paths": [ | |
| "pocket/model_fp32.onnx", | |
| "pocket/model.pt" | |
| ], | |
| "vocab": "pocket/vocab.json", | |
| "architecture": "omniASR_CTC_300M_v2 student distilled from a 1B Dynamic ASR teacher", | |
| "runtime": "onnxruntime CPU, greedy CTC, 16 kHz mono", | |
| "script": "pocket/transcribe_pocket.py", | |
| "results": "pocket/results.json", | |
| "status": "CPU variant" | |
| } | |
| }, | |
| "decoding": { | |
| "type": "greedy CTC", | |
| "latin_script_mask": true, | |
| "language_hint": null, | |
| "streaming": "re-decode the growing buffer (duplex asr_server)" | |
| }, | |
| "output": "lowercase, no punctuation; Igbo without tone marks (dot letters kept); Yoruba with tone marks", | |
| "weights_license": "cc-by-sa-4.0", | |
| "code_license": "apache-2.0", | |
| "scores": { | |
| "v1_stock": { | |
| "duplex_eng": { | |
| "n": 29, | |
| "wer": 0.1437, | |
| "cer": 0.0654, | |
| "cer_nodiac": 0.0654 | |
| }, | |
| "duplex_pcm": { | |
| "n": 40, | |
| "wer": 0.3796, | |
| "cer": 0.1693, | |
| "cer_nodiac": 0.169 | |
| }, | |
| "duplex_ibo": { | |
| "n": 40, | |
| "wer": 0.636, | |
| "cer": 0.2057, | |
| "cer_nodiac": 0.1614 | |
| }, | |
| "duplex_yor": { | |
| "n": 40, | |
| "wer": 0.4978, | |
| "cer": 0.1905, | |
| "cer_nodiac": 0.1227 | |
| }, | |
| "duplex_hau": { | |
| "n": 40, | |
| "wer": 0.299, | |
| "cer": 0.0897, | |
| "cer_nodiac": 0.0885 | |
| }, | |
| "duplex_cs": { | |
| "n": 40, | |
| "wer": 0.3295, | |
| "cer": 0.1282, | |
| "cer_nodiac": 0.1053 | |
| }, | |
| "realcs_all": { | |
| "n": 94, | |
| "wer": 0.6062, | |
| "cer": 0.2243, | |
| "cer_nodiac": 0.1951 | |
| }, | |
| "realcs_natural_conversational": { | |
| "n": 36, | |
| "wer": 0.8351, | |
| "cer": 0.3719, | |
| "cer_nodiac": 0.3256 | |
| }, | |
| "realcs_insertional": { | |
| "n": 58, | |
| "wer": 0.5195, | |
| "cer": 0.1783, | |
| "cer_nodiac": 0.1544 | |
| }, | |
| "realcs_eng_word_recall": 0.3009, | |
| "synthtest": { | |
| "n": 109, | |
| "wer": 0.4141, | |
| "cer": 0.1586, | |
| "cer_nodiac": 0.1293 | |
| } | |
| }, | |
| "v2_stock": { | |
| "duplex_eng": { | |
| "n": 29, | |
| "wer": 0.0969, | |
| "cer": 0.0244, | |
| "cer_nodiac": 0.0244 | |
| }, | |
| "duplex_pcm": { | |
| "n": 40, | |
| "wer": 0.3358, | |
| "cer": 0.1232, | |
| "cer_nodiac": 0.1223 | |
| }, | |
| "duplex_ibo": { | |
| "n": 40, | |
| "wer": 0.4559, | |
| "cer": 0.1329, | |
| "cer_nodiac": 0.1089 | |
| }, | |
| "duplex_yor": { | |
| "n": 40, | |
| "wer": 0.4009, | |
| "cer": 0.1346, | |
| "cer_nodiac": 0.0977 | |
| }, | |
| "duplex_hau": { | |
| "n": 40, | |
| "wer": 0.3172, | |
| "cer": 0.1061, | |
| "cer_nodiac": 0.1059 | |
| }, | |
| "duplex_cs": { | |
| "n": 40, | |
| "wer": 0.2922, | |
| "cer": 0.1042, | |
| "cer_nodiac": 0.0898 | |
| }, | |
| "realcs_all": { | |
| "n": 94, | |
| "wer": 0.5139, | |
| "cer": 0.1931, | |
| "cer_nodiac": 0.1725 | |
| }, | |
| "realcs_natural_conversational": { | |
| "n": 36, | |
| "wer": 0.7773, | |
| "cer": 0.3535, | |
| "cer_nodiac": 0.3152 | |
| }, | |
| "realcs_insertional": { | |
| "n": 58, | |
| "wer": 0.4141, | |
| "cer": 0.1431, | |
| "cer_nodiac": 0.1281 | |
| }, | |
| "realcs_eng_word_recall": 0.2832, | |
| "synthtest": { | |
| "n": 109, | |
| "wer": 0.3541, | |
| "cer": 0.1324, | |
| "cer_nodiac": 0.1179 | |
| } | |
| }, | |
| "A_real": { | |
| "duplex_eng": { | |
| "n": 29, | |
| "wer": 0.0953, | |
| "cer": 0.0252, | |
| "cer_nodiac": 0.0252 | |
| }, | |
| "duplex_pcm": { | |
| "n": 40, | |
| "wer": 0.1927, | |
| "cer": 0.0738, | |
| "cer_nodiac": 0.0736 | |
| }, | |
| "duplex_ibo": { | |
| "n": 40, | |
| "wer": 0.3153, | |
| "cer": 0.0929, | |
| "cer_nodiac": 0.0761 | |
| }, | |
| "duplex_yor": { | |
| "n": 40, | |
| "wer": 0.3547, | |
| "cer": 0.1213, | |
| "cer_nodiac": 0.0841 | |
| }, | |
| "duplex_hau": { | |
| "n": 40, | |
| "wer": 0.2533, | |
| "cer": 0.0823, | |
| "cer_nodiac": 0.0823 | |
| }, | |
| "duplex_cs": { | |
| "n": 40, | |
| "wer": 0.2385, | |
| "cer": 0.0807, | |
| "cer_nodiac": 0.0646 | |
| }, | |
| "realcs_all": { | |
| "n": 94, | |
| "wer": 0.4357, | |
| "cer": 0.1482, | |
| "cer_nodiac": 0.137 | |
| }, | |
| "realcs_natural_conversational": { | |
| "n": 36, | |
| "wer": 0.6227, | |
| "cer": 0.2481, | |
| "cer_nodiac": 0.2335 | |
| }, | |
| "realcs_insertional": { | |
| "n": 58, | |
| "wer": 0.3648, | |
| "cer": 0.117, | |
| "cer_nodiac": 0.107 | |
| }, | |
| "realcs_eng_word_recall": 0.3186, | |
| "synthtest": { | |
| "n": 109, | |
| "wer": 0.2932, | |
| "cer": 0.1052, | |
| "cer_nodiac": 0.0914 | |
| } | |
| }, | |
| "B_synth": { | |
| "duplex_eng": { | |
| "n": 29, | |
| "wer": 0.0922, | |
| "cer": 0.0241, | |
| "cer_nodiac": 0.0241 | |
| }, | |
| "duplex_pcm": { | |
| "n": 40, | |
| "wer": 0.1927, | |
| "cer": 0.0761, | |
| "cer_nodiac": 0.0759 | |
| }, | |
| "duplex_ibo": { | |
| "n": 40, | |
| "wer": 0.3207, | |
| "cer": 0.095, | |
| "cer_nodiac": 0.0754 | |
| }, | |
| "duplex_yor": { | |
| "n": 40, | |
| "wer": 0.3532, | |
| "cer": 0.1182, | |
| "cer_nodiac": 0.0782 | |
| }, | |
| "duplex_hau": { | |
| "n": 40, | |
| "wer": 0.252, | |
| "cer": 0.0808, | |
| "cer_nodiac": 0.0808 | |
| }, | |
| "duplex_cs": { | |
| "n": 40, | |
| "wer": 0.2157, | |
| "cer": 0.0718, | |
| "cer_nodiac": 0.0648 | |
| }, | |
| "realcs_all": { | |
| "n": 94, | |
| "wer": 0.438, | |
| "cer": 0.1492, | |
| "cer_nodiac": 0.1368 | |
| }, | |
| "realcs_natural_conversational": { | |
| "n": 36, | |
| "wer": 0.6103, | |
| "cer": 0.2472, | |
| "cer_nodiac": 0.2339 | |
| }, | |
| "realcs_insertional": { | |
| "n": 58, | |
| "wer": 0.3727, | |
| "cer": 0.1186, | |
| "cer_nodiac": 0.1065 | |
| }, | |
| "realcs_eng_word_recall": 0.3805, | |
| "synthtest": { | |
| "n": 109, | |
| "wer": 0.1261, | |
| "cer": 0.0385, | |
| "cer_nodiac": 0.0349 | |
| } | |
| } | |
| }, | |
| "excluded_data": [ | |
| "African Voices / WaZoBia", | |
| "NaijaVoices (non-commercial)", | |
| "AfriSpeech-200 (non-commercial)", | |
| "Common Voice (not fetched)" | |
| ] | |
| } |