fix lint

egs/aishell/ASR/README.md

@@ -24,3 +24,10 @@ The following table lists the differences among them.
 The decoder in `transducer_stateless` is modified from the paper
 [Rnn-Transducer with Stateless Prediction Network](https://ieeexplore.ieee.org/document/9054419/).
 We place an additional Conv1d layer right after the input embedding layer.
+
+# Whisper
+
+Recipe to finetune large pretrained models
+|                                    | Encoder   | Decoder            | Comment                                                                           |
+|------------------------------------|-----------|--------------------|-----------------------------------------------------------------------------------|
+| `whisper`             | Transformer | Transformer | support fine-tuning using deepspeed

egs/aishell/ASR/RESULTS.md

@@ -77,7 +77,7 @@ It's reworked Zipformer with Pruned RNNT loss.
 
 Command for training is:
 ```bash
-./prepare.sh 
+./prepare.sh
 
 export CUDA_VISIBLE_DEVICES="0,1"
 
@@ -142,7 +142,7 @@ export CUDA_VISIBLE_DEVICES="0,1"
   --feedforward-dim 512,768,768,768,768,768 \
   --encoder-dim 192,256,256,256,256,256 \
   --encoder-unmasked-dim 192,192,192,192,192,192 \
-  --max-duration 1200 
+  --max-duration 1200
 ```
 
 Command for decoding is:
@@ -192,7 +192,7 @@ export CUDA_VISIBLE_DEVICES="0,1"
   --feedforward-dim 512,768,1536,2048,1536,768 \
   --encoder-dim 192,256,512,768,512,256 \
   --encoder-unmasked-dim 192,192,256,320,256,192 \
-  --max-duration 800 
+  --max-duration 800
 ```
 
 Command for decoding is:
@@ -208,7 +208,7 @@ for m in greedy_search modified_beam_search fast_beam_search ; do
     --num-encoder-layers 2,2,4,5,4,2 \
     --feedforward-dim 512,768,1536,2048,1536,768 \
     --encoder-dim 192,256,512,768,512,256 \
-    --encoder-unmasked-dim 192,192,256,320,256,192 
+    --encoder-unmasked-dim 192,192,256,320,256,192
 done
 ```
 

egs/aishell/ASR/local/compute_fbank_aishell.py

@@ -29,7 +29,14 @@
 from pathlib import Path
 
 import torch
-from lhotse import CutSet, Fbank, FbankConfig, WhisperFbank, WhisperFbankConfig, LilcomChunkyWriter
+from lhotse import (
+    CutSet,
+    Fbank,
+    FbankConfig,
+    LilcomChunkyWriter,
+    WhisperFbank,
+    WhisperFbankConfig,
+)
 from lhotse.recipes.utils import read_manifests_if_cached
 
 from icefall.utils import get_executor, str2bool
@@ -42,7 +49,9 @@
 torch.set_num_interop_threads(1)
 
 
-def compute_fbank_aishell(num_mel_bins: int = 80, perturb_speed: bool = False, whisper_fbank: bool = False):
+def compute_fbank_aishell(
+    num_mel_bins: int = 80, perturb_speed: bool = False, whisper_fbank: bool = False
+):
     src_dir = Path("data/manifests")
     output_dir = Path("data/fbank")
     num_jobs = min(15, os.cpu_count())
@@ -69,7 +78,9 @@ def compute_fbank_aishell(num_mel_bins: int = 80, perturb_speed: bool = False, w
         dataset_parts,
     )
     if whisper_fbank:
-        extractor = WhisperFbank(WhisperFbankConfig(num_filters=num_mel_bins, device='cuda'))
+        extractor = WhisperFbank(
+            WhisperFbankConfig(num_filters=num_mel_bins, device="cuda")
+        )
     else:
         extractor = Fbank(FbankConfig(num_mel_bins=num_mel_bins))
 
@@ -84,7 +95,7 @@ def compute_fbank_aishell(num_mel_bins: int = 80, perturb_speed: bool = False, w
                 supervisions=m["supervisions"],
             )
             if "train" in partition and 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)
                 )
@@ -129,5 +140,7 @@ def get_args():
 
     args = get_args()
     compute_fbank_aishell(
-        num_mel_bins=args.num_mel_bins, perturb_speed=args.perturb_speed, whisper_fbank=args.whisper_fbank
+        num_mel_bins=args.num_mel_bins,
+        perturb_speed=args.perturb_speed,
+        whisper_fbank=args.whisper_fbank,
     )

egs/aishell/ASR/prepare.sh

@@ -387,4 +387,4 @@ if [ $stage -le 30 ] && [ $stop_stage -ge 30 ]; then
     ./local/compute_fbank_musan.py --num-mel-bins ${whisper_mel_bins} --whisper-fbank true
     touch data/fbank/.aishell.whisper.done
   fi
-fi
+fi