Merge 1775f36a612b4bbc351077d51c4e40540678d61f into 34fc1fdf0d8ff520e2bb18267d046ca207c78ef9

egs/librispeech/ASR/local/compute_fbank_librispeech.py

@@ -68,6 +68,13 @@ def get_args():
         help="""Perturb speed with factor 0.9 and 1.1 on train subset.""",
     )
 
+    parser.add_argument(
+        "--num-workers",
+        type=int,
+        default=15,
+        help="Number of worker processes for feature extraction.",
+    )
+
     return parser.parse_args()
 
 
@@ -75,10 +82,11 @@ def compute_fbank_librispeech(
     bpe_model: Optional[str] = None,
     dataset: Optional[str] = None,
     perturb_speed: Optional[bool] = True,
+    num_workers: int = 15,
 ):
     src_dir = Path("data/manifests")
     output_dir = Path("data/fbank")
-    num_jobs = min(15, os.cpu_count())
+    num_jobs = min(num_workers, os.cpu_count())
     num_mel_bins = 80
 
     if bpe_model:
@@ -125,6 +133,7 @@ def compute_fbank_librispeech(
                 logging.info(f"{partition} already exists - skipping.")
                 continue
             logging.info(f"Processing {partition}")
+
             cut_set = CutSet.from_manifests(
                 recordings=m["recordings"],
                 supervisions=m["supervisions"],
@@ -134,20 +143,44 @@ def compute_fbank_librispeech(
                 if bpe_model:
                     cut_set = filter_cuts(cut_set, sp)
                 if perturb_speed:
-                    logging.info(f"Doing speed perturb")
+                    logging.info("Doing speed perturb")
                     cut_set = (
                         cut_set
                         + cut_set.perturb_speed(0.9)
                         + cut_set.perturb_speed(1.1)
                     )
+
+            if ex is None:
+                # Create a custom process pool context for None (local execution)
+                import multiprocessing as mp
+                from concurrent.futures import ProcessPoolExecutor
+
+                # Calculate the number of jobs
+                actual_jobs = (
+                    min(num_jobs * 2, 20) if "train" in partition else num_jobs
+                )
+
+                # Use the forkserver method
+                ctx = mp.get_context("forkserver")
+                with ProcessPoolExecutor(
+                    max_workers=actual_jobs, mp_context=ctx
+                ) as local_executor:
                     cut_set = cut_set.compute_and_store_features(
                         extractor=extractor,
                         storage_path=f"{output_dir}/{prefix}_feats_{partition}",
-                # when an executor is specified, make more partitions
+                        executor=local_executor,
-                num_jobs=num_jobs if ex is None else 80,
+                        storage_type=LilcomChunkyWriter,
+                    )
+            else:
+                # Distributed environment, use the provided executor
+                cut_set = cut_set.compute_and_store_features(
+                    extractor=extractor,
+                    storage_path=f"{output_dir}/{prefix}_feats_{partition}",
+                    num_jobs=min(num_jobs * 2, 20),
                     executor=ex,
                     storage_type=LilcomChunkyWriter,
                 )
+
             cut_set.to_file(output_dir / cuts_filename)
 
 
@@ -161,4 +194,5 @@ if __name__ == "__main__":
         bpe_model=args.bpe_model,
         dataset=args.dataset,
         perturb_speed=args.perturb_speed,
+        num_workers=args.num_workers,
     )

egs/librispeech/ASR/prepare.sh

@@ -139,7 +139,7 @@ if [ $stage -le 3 ] && [ $stop_stage -ge 3 ]; then
   log "Stage 3: Compute fbank for librispeech"
   mkdir -p data/fbank
   if [ ! -e data/fbank/.librispeech.done ]; then
-    ./local/compute_fbank_librispeech.py
+    ./local/compute_fbank_librispeech.py --num-workers $nj
     touch data/fbank/.librispeech.done
   fi
 

icefall/utils.py

@@ -180,10 +180,10 @@ def setup_logger(
         world_size = dist.get_world_size()
         rank = dist.get_rank()
         formatter = f"%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] ({rank}/{world_size}) %(message)s"  # noqa
-        log_filename = f"{log_filename}-{date_time}-{rank}"
+        log_filename = f"{log_filename}-{date_time}-{rank}.log"
     else:
         formatter = "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
-        log_filename = f"{log_filename}-{date_time}"
+        log_filename = f"{log_filename}-{date_time}.log"
 
     os.makedirs(os.path.dirname(log_filename), exist_ok=True)