Merge branch 'main' of https://huggingface.co/osanseviero/asr-with-transformers-wav2vec2 into main

README.md

@@ -1,10 +1,13 @@
 ---
+benchmark: superb
+library_name: superb
 language: en
 datasets:
 - librispeech_asr
 tags:
 - audio
 - automatic-speech-recognition
+- superb
 license: apache-2.0
 widget:
 - label: Librispeech sample 1

model.py

@@ -3,7 +3,6 @@ from transformers import AutomaticSpeechRecognitionPipeline, AutoTokenizer, Wav2
 from typing import Dict
 from pathlib import Path
 
-
 class PreTrainedModel():
     def __init__(self):
         """
@@ -26,17 +25,12 @@ class PreTrainedModel():
         """
         return self.model(inputs)
 
-
 """
 # Just an example using this.
 import subprocess
 from datasets import load_dataset
 
 def ffmpeg_read(bpayload: bytes, sampling_rate: int) -> np.array:
-    """
-    Librosa does that under the hood but forces the use of an actual
-    file leading to hitting disk, which is almost always very bad.
-    """
     ar = f"{sampling_rate}"
     ac = "1"
     format_for_conversion = "f32le"

requirements.txt

@@ -1,2 +1,3 @@
 transformers
-numpy
+numpy
+torch