From 1ef50cdbc2d229d54bc611e54b8b2804118e3325 Mon Sep 17 00:00:00 2001 From: Taejin Park Date: Fri, 9 Feb 2024 15:01:44 -0800 Subject: [PATCH] Merged the latest nemo main Signed-off-by: Taejin Park --- nemo/collections/asr/metrics/der.py | 10 ++++------ nemo/collections/asr/models/msdd_v2_models.py | 3 +-- scripts/chime7/pipeline/asr/run_asr.py | 4 ++-- 3 files changed, 7 insertions(+), 10 deletions(-) diff --git a/nemo/collections/asr/metrics/der.py b/nemo/collections/asr/metrics/der.py index e1a0fde45d46..d4e68c0effbc 100644 --- a/nemo/collections/asr/metrics/der.py +++ b/nemo/collections/asr/metrics/der.py @@ -182,13 +182,11 @@ def score_labels( itemized_errors = (DER, CER, FA, MISS) if verbose: - # logging.info(f"\n{metric.report()}") - pass - logging.info( - "Cumulative Results for collar {} sec and ignore_overlap {}: \n| FA: {:.4f} | MISS: {:.4f} | CER: {:.4f} | DER: {:.4f} | Spk. Count Acc. {:.4f}\n".format( - collar, ignore_overlap, FA, MISS, CER, DER, spk_count_acc + logging.info( + "Cumulative Results for collar {} sec and ignore_overlap {}: \n| FA: {:.4f} | MISS: {:.4f} | CER: {:.4f} | DER: {:.4f} | Spk. Count Acc. {:.4f}\n".format( + collar, ignore_overlap, FA, MISS, CER, DER, spk_count_acc + ) ) - ) return metric, mapping_dict, itemized_errors elif verbose: diff --git a/nemo/collections/asr/models/msdd_v2_models.py b/nemo/collections/asr/models/msdd_v2_models.py index a2374a60faf2..04b059edd67a 100644 --- a/nemo/collections/asr/models/msdd_v2_models.py +++ b/nemo/collections/asr/models/msdd_v2_models.py @@ -1669,7 +1669,7 @@ def __init__(self, cfg: Union[DictConfig, NeuralDiarizerInferenceConfig], msdd_m self.use_adaptive_thres = cfg.diarizer.msdd_model.parameters.get('use_adaptive_thres', True) self.max_pred_length = cfg.diarizer.msdd_model.parameters.get('max_pred_length', 0) self.diar_eval_settings = cfg.diarizer.msdd_model.parameters.get( - 'diar_eval_settings', [(0.25, False), (0.25, True)] + 'diar_eval_settings', [(0.25, False)] ) if msdd_model is not None: self.msdd_model = msdd_model @@ -2164,7 +2164,6 @@ def run_overlap_aware_eval( hop_len_in_cs=int(self.feat_per_sec * self.msdd_model.cfg.interpolated_scale/2), ts_vad_threshold=self._cfg.diarizer.msdd_model.parameters.ts_vad_threshold, ) - for k, (collar, ignore_overlap) in enumerate(self.diar_eval_settings): output = score_labels( rttm_map, diff --git a/scripts/chime7/pipeline/asr/run_asr.py b/scripts/chime7/pipeline/asr/run_asr.py index 4cd51dd6ab6f..2f3bd7c6453f 100644 --- a/scripts/chime7/pipeline/asr/run_asr.py +++ b/scripts/chime7/pipeline/asr/run_asr.py @@ -267,13 +267,13 @@ def autocast(): ) else: transcriptions = asr_model.transcribe( - paths2audio_files=filepaths, + audio=filepaths, batch_size=cfg.batch_size, num_workers=cfg.num_workers, return_hypotheses=return_hypotheses, channel_selector=cfg.channel_selector, augmentor=augmentor, - normalize_db=cfg.normalize_db, + # normalize_db=cfg.normalize_db, ) logging.info(f"Finished transcribing {len(filepaths)} files !")