[ { "model": "Vosk", "repo": "https://alphacephei.com/vosk/models/vosk-model-fa-0.42.zip", "family": "Vosk / Kaldi", "precision": "fp32", "params_b": 0.494933674, "gold69_wer": 43.33740831295844, "gold69_cer": 22.621107266435985, "fleurs_wer": 13.1608339538347, "fleurs_cer": 5.782299678707039, "mean_decode_ms": 3338.437947674074, "status": "complete", "notes": "Corrected Gold69 v2 rerun completed on Stallion with the base unquantized model. FLEURS is from the earlier full double eval.", "gold69_v2_fair_wer": 41.01157830591103, "gold69_v2_fair_cer": 22.467942761568484, "gold69_v2_revision": "09f80c368937e8899c952774776d3e54cf49768c", "gold69_v2_n": 69 }, { "model": "nezamisafa/whisper-persian-v4", "repo": "https://huggingface.co/nezamisafa/whisper-persian-v4", "family": "Whisper Large v3", "precision": "fp32", "params_b": 1.54349056, "gold69_wer": 34.84107579462103, "gold69_cer": 14.749134948096888, "fleurs_wer": 13.212025316455698, "fleurs_cer": 4.992697887255379, "mean_decode_ms": 553.3734853716865, "status": "complete", "notes": "Corrected Gold69 v2 rerun completed on Stallion with the base unquantized model. FLEURS is from the earlier full double eval.", "gold69_v2_fair_wer": 31.68799512492383, "gold69_v2_fair_cer": 12.562720683887754, "gold69_v2_revision": "09f80c368937e8899c952774776d3e54cf49768c", "gold69_v2_n": 69 }, { "model": "VisualEars FastConformer full A+B CTC + 4gram LM", "repo": "local:/workspace/visualears_full_pipeline/training/fastconformer_fa_full_ab/runs/stt_fa_fastconformer_hybrid_large/fa_fastconformer_ab_final.nemo", "family": "NVIDIA FastConformer Hybrid CTC + 4-gram LM", "precision": "bf16 train / fp32 eval", "params_b": 0.114621442, "gold69_v2_fair_wer": 25.95978062157221, "gold69_v2_fair_cer": 12.61847240289909, "gold69_v2_n": 69, "gold69_wer": 25.95978062157221, "gold69_cer": 12.61847240289909, "fleurs_wer": 13.602941176470587, "fleurs_cer": 5.394990366088632, "fleurs_n": 871, "mean_decode_ms": 18.27165043039245, "status": "complete", "notes": "Same full A+B FastConformer final .nemo decoded through the CTC head with the train-split 4-gram LM. FLEURS-256 calibrated alpha=0.2 beta=-1.0 beam=50; improves FLEURS WER but hurts Golha CER versus RNNT greedy." }, { "model": "VisualEars v2 final + 4gram LM", "repo": "https://huggingface.co/Reza2kn/visualears-xlsr-300m-v2-cont-step1568", "family": "XLS-R / Wav2Vec2 CTC + 4-gram LM", "params_b": 0.315511495, "gold69_v2_fair_wer": 22.91285801340646, "gold69_v2_fair_cer": 9.19903363687047, "gold69_v2_revision": "09f80c368937e8899c952774776d3e54cf49768c", "gold69_v2_n": 69, "gold69_wer": 25.68976088289393, "gold69_cer": 10.164597170083741, "fleurs_wer": 14.26546330860464, "fleurs_cer": 5.101900842641583, "mean_decode_ms": 151.5814843385116, "status": "complete", "notes": "Same final checkpoint with train-split 4-gram LM rescoring. Word score improves; fair CER worsens versus greedy." }, { "model": "vhdm/whisper-large-fa-v1", "repo": "https://huggingface.co/vhdm/whisper-large-fa-v1", "family": "Whisper Large v3 Turbo", "precision": "fp32", "params_b": 0.80887808, "gold69_wer": 45.72127139364303, "gold69_cer": 24.134948096885815, "fleurs_wer": 14.663998510796725, "fleurs_cer": 5.576866906825041, "mean_decode_ms": 172.2892229107843, "status": "complete", "notes": "Corrected Gold69 v2 rerun completed on Stallion with the base unquantized model. FLEURS is from the earlier full double eval.", "gold69_v2_fair_wer": 43.69287020109689, "gold69_v2_fair_cer": 21.70600260174689, "gold69_v2_revision": "09f80c368937e8899c952774776d3e54cf49768c", "gold69_v2_n": 69 }, { "model": "VisualEars FastConformer full A+B final", "repo": "local:/workspace/visualears_full_pipeline/training/fastconformer_fa_full_ab/runs/stt_fa_fastconformer_hybrid_large/fa_fastconformer_ab_final.nemo", "family": "NVIDIA FastConformer Hybrid RNNT/CTC", "precision": "bf16 train / fp32 eval", "params_b": 0.114621442, "gold69_v2_fair_wer": 25.28945764777575, "gold69_v2_fair_cer": 9.64504738896116, "gold69_v2_n": 69, "gold69_wer": 25.28945764777575, "gold69_cer": 9.64504738896116, "fleurs_wer": 15.734363365599405, "fleurs_cer": 5.246213506987635, "fleurs_n": 871, "mean_decode_ms": 9.34309663783532, "status": "complete", "notes": "Full A+B 9.19k-hour FastConformer final .nemo. Golha uses cached gold69 v2 fair scoring; FLEURS uses upstream google/fleurs fa_ir test with same normalization, including diacritic stripping. No LM rescoring." }, { "model": "VisualEars v2 final greedy", "repo": "https://huggingface.co/Reza2kn/visualears-xlsr-300m-v2-cont-step1568", "family": "XLS-R / Wav2Vec2 CTC", "params_b": 0.315511495, "gold69_v2_fair_wer": 25.167580743449115, "gold69_v2_fair_cer": 7.693737223564394, "gold69_v2_revision": "09f80c368937e8899c952774776d3e54cf49768c", "gold69_v2_n": 69, "gold69_wer": 27.71305947271612, "gold69_cer": 9.572624891712389, "fleurs_wer": 19.132448140622785, "fleurs_cer": 5.875955320399765, "mean_decode_ms": 119.35669097347537, "status": "complete", "notes": "Final VisualEars v2 continuation, greedy CTC decode. Corrected Gold69 v2 fair scoring added." }, { "model": "student_mms1b", "repo": "local:/home/rezo/golha/student_mms1b_eval", "family": "MMS / Wav2Vec2 CTC", "precision": "fp32", "params_b": 0.965, "gold69_wer": 41.62591687041565, "gold69_cer": 14.316608996539792, "fleurs_wer": 19.592330603127326, "fleurs_cer": 5.239022490507254, "mean_decode_ms": 186.91, "status": "complete", "notes": "Existing saved predictions rescored against corrected Gold69 v2; FLEURS unchanged from previous double eval.", "gold69_v2_fair_wer": 39.97562461913468, "gold69_v2_fair_cer": 11.652109273369263, "gold69_v2_revision": "09f80c368937e8899c952774776d3e54cf49768c", "gold69_v2_n": 69 }, { "model": "Peacockery/omni-ctc-300m-v2-fleurs-fa-ir-thomcles-continue", "repo": "https://huggingface.co/Peacockery/omni-ctc-300m-v2-fleurs-fa-ir-thomcles-continue", "family": "Omnilingual ASR CTC", "precision": "fp32", "params_b": 0.325982896, "gold69_wer": 45.11002444987775, "gold69_cer": 14.677047289504037, "fleurs_wer": 19.59233060312733, "fleurs_cer": 5.2867296271054425, "mean_decode_ms": 13.531180548329514, "status": "complete", "notes": "Existing saved predictions rescored against corrected Gold69 v2; FLEURS unchanged from previous double eval.", "gold69_v2_fair_wer": 41.98659354052407, "gold69_v2_fair_cer": 11.875116149414607, "gold69_v2_revision": "09f80c368937e8899c952774776d3e54cf49768c", "gold69_v2_n": 69 }, { "model": "VisualEars v2 checkpoint-1000", "repo": "https://huggingface.co/Reza2kn/visualears-xlsr-300m-v2-cont-step1568", "family": "XLS-R / Wav2Vec2 CTC", "params_b": 0.315511495, "gold69_v2_fair_wer": 26.508226691042047, "gold69_v2_fair_cer": 7.65656941089017, "gold69_v2_revision": "09f80c368937e8899c952774776d3e54cf49768c", "gold69_v2_n": 69, "gold69_wer": 28.973105134474324, "gold69_cer": 9.77508650519031, "fleurs_wer": 19.98324646314222, "fleurs_cer": 6.292473955797878, "mean_decode_ms": 31.10736003820447, "status": "complete", "notes": "Diagnostic earlier checkpoint from the VisualEars v2 run." }, { "model": "student_xlsr_large_relabel", "repo": "https://huggingface.co/Reza2kn/golha-xlsr-300m-persian-asr-large-relabel", "family": "XLS-R / Wav2Vec2 CTC", "precision": "fp32", "params_b": 0.315511495, "gold69_wer": 28.85085574572127, "gold69_cer": 9.674163783160324, "fleurs_wer": 20.309009679821298, "fleurs_cer": 5.764774608119949, "mean_decode_ms": 109.25901333491008, "status": "complete", "notes": "Previous best rerun on corrected Gold69 v2. FLEURS is from the earlier full double eval.", "gold69_v2_fair_wer": 26.508226691042047, "gold69_v2_fair_cer": 7.452146441181936, "gold69_v2_revision": "09f80c368937e8899c952774776d3e54cf49768c", "gold69_v2_n": 69 }, { "model": "Qwen/Qwen3-ASR-1.7B", "repo": "https://huggingface.co/Qwen/Qwen3-ASR-1.7B", "family": "Qwen3 ASR", "precision": "bf16", "params_b": 2.03805248, "gold69_wer": 62.28606356968215, "gold69_cer": 27.03287197231834, "fleurs_wer": 24.604430379746837, "fleurs_cer": 8.778113134066789, "mean_decode_ms": 892.0253912607828, "status": "complete", "notes": "Corrected Gold69 v2 rerun completed on Stallion with the base unquantized model. FLEURS is from the earlier full double eval.", "gold69_v2_fair_wer": 60.3290676416819, "gold69_v2_fair_cer": 25.59003902620331, "gold69_v2_revision": "09f80c368937e8899c952774776d3e54cf49768c", "gold69_v2_n": 69 }, { "model": "nvidia/stt_fa_fastconformer_hybrid_large", "repo": "https://huggingface.co/nvidia/stt_fa_fastconformer_hybrid_large", "family": "NVIDIA FastConformer Hybrid", "precision": "fp32", "params_b": 0.114621442, "gold69_wer": 73.10513447432763, "gold69_cer": 55.14705882352941, "gold69_v2_fair_wer": 53.56489945155393, "gold69_v2_fair_cer": 25.64579074521464, "gold69_v2_revision": "09f80c368937e8899c952774776d3e54cf49768c", "gold69_v2_n": 69, "fleurs_wer": 45.00651526433358, "fleurs_cer": 26.84451367929121, "mean_decode_ms": 18.27055129452028, "status": "complete", "notes": "Gold69 v2 rerun with standardized fair scorer on full-precision local NeMo checkpoint." }, { "model": "Qwen/Qwen3-ASR-0.6B", "repo": "https://huggingface.co/Qwen/Qwen3-ASR-0.6B", "family": "Qwen3 ASR", "precision": "bf16", "params_b": 0.782426112, "gold69_wer": 76.7114914425428, "gold69_cer": 36.60611303344867, "fleurs_wer": 46.89128816083395, "fleurs_cer": 19.847142439879274, "mean_decode_ms": 662.3744273531264, "status": "complete", "notes": "Corrected Gold69 v2 rerun completed on Stallion with the base unquantized model. FLEURS is from the earlier full double eval.", "gold69_v2_fair_wer": 76.23400365630712, "gold69_v2_fair_cer": 36.35012079539119, "gold69_v2_revision": "09f80c368937e8899c952774776d3e54cf49768c", "gold69_v2_n": 69 }, { "model": "microsoft/VibeVoice-ASR", "repo": "https://huggingface.co/microsoft/VibeVoice-ASR", "family": "VibeVoice ASR", "precision": "bf16", "params_b": 8.674, "gold69_wer": 76.83374083129584, "gold69_cer": 49.942329873125715, "fleurs_wer": 29.27215189873418, "fleurs_cer": 15.438613572193555, "mean_decode_ms": 1737.57, "status": "gold69 v2 pending", "notes": "FLEURS shown from previous public-baseline run. Corrected Gold69 v2 rerun is pending, so stale Gold69 v1 numbers are not used for ranking." } ]