Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -21,7 +21,6 @@ configs:
|
|
| 21 |
license: apache-2.0
|
| 22 |
task_categories:
|
| 23 |
- automatic-speech-recognition
|
| 24 |
-
- translation
|
| 25 |
language:
|
| 26 |
- ja
|
| 27 |
tags:
|
|
@@ -30,4 +29,53 @@ tags:
|
|
| 30 |
- asr
|
| 31 |
size_categories:
|
| 32 |
- 10K<n<100K
|
| 33 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
license: apache-2.0
|
| 22 |
task_categories:
|
| 23 |
- automatic-speech-recognition
|
|
|
|
| 24 |
language:
|
| 25 |
- ja
|
| 26 |
tags:
|
|
|
|
| 29 |
- asr
|
| 30 |
size_categories:
|
| 31 |
- 10K<n<100K
|
| 32 |
+
---
|
| 33 |
+
|
| 34 |
+
<pre><code>
|
| 35 |
+
# Extracts audio files and transcriptions from datasets. Creates an mp3/wav and a text file with transcription for each sample.
|
| 36 |
+
|
| 37 |
+
import soundfile as sf, os, csv, shutil
|
| 38 |
+
from datasets import load_dataset
|
| 39 |
+
from tqdm import tqdm
|
| 40 |
+
|
| 41 |
+
dataset = load_dataset("dataset", split="train", trust_remote_code=True, streaming=True)
|
| 42 |
+
|
| 43 |
+
name = "dataset"
|
| 44 |
+
os.makedirs("./datasets/" + name, exist_ok=True) # Create a folder to store the audio and transcription files
|
| 45 |
+
folder_path = "./datasets/" + name
|
| 46 |
+
filter = dataset.filter(lambda sample: sample["language"] == "Japanese") # Filter out samples you do/don't want
|
| 47 |
+
|
| 48 |
+
# This creates the audio and transcripition files and saves each individually
|
| 49 |
+
|
| 50 |
+
for i, sample in tqdm(enumerate(dataset)): # Process each sample in the dataset - use filter if filtering
|
| 51 |
+
audio_path = os.path.join(folder_path, name + f'_{i}.mp3') # Path to save the audio file - can change mp3 to other format
|
| 52 |
+
transcription_path = os.path.join(folder_path, name + f'_{i}.txt') # Path to save the transcription file
|
| 53 |
+
sf.write(audio_path, sample['audio']['array'], sample['audio']['sampling_rate']) # Save the audio file
|
| 54 |
+
sample["audio_length"] = len(sample["audio"]["array"]) / sample["audio"]["sampling_rate"] # Get audio length
|
| 55 |
+
with open(transcription_path, 'w', encoding='utf-8') as transcription_file: # Save the transcription file
|
| 56 |
+
transcription_file.write(sample['sentence']) # Save the transcription sentence
|
| 57 |
+
transcription_file.write(str(","+str(sample['audio_length']))) # Save the audio length
|
| 58 |
+
|
| 59 |
+
# This creates a metadata.csv file with file_names and sentences in columns and then moves transcription text files to txt directory
|
| 60 |
+
|
| 61 |
+
name = name
|
| 62 |
+
folder_path = "./datasets/" + name
|
| 63 |
+
output_file = folder_path + '/metadata.csv'
|
| 64 |
+
|
| 65 |
+
with open(output_file, 'w', encoding='utf-8') as outfile:
|
| 66 |
+
writer = csv.writer(outfile) # Creates csv
|
| 67 |
+
writer.writerow(('file_name', 'sentence')) # Creates column heads
|
| 68 |
+
for file_name in os.listdir(folder_path):
|
| 69 |
+
if file_name.endswith('.txt'):
|
| 70 |
+
filepath = os.path.join(folder_path, file_name)
|
| 71 |
+
with open(filepath, 'r', encoding='utf-8') as infile:
|
| 72 |
+
outfile.write((file_name.rstrip('.txt'))+".mp3" + ",") # Add file names with trailing comma to csv
|
| 73 |
+
outfile.write(infile.read()) # Add sentence/transcriptions on the same line as file name
|
| 74 |
+
outfile.write('\n') # Adds a return
|
| 75 |
+
for file_name in os.listdir(folder_path):
|
| 76 |
+
os.makedirs(folder_path + "./txt/", exist_ok=True)
|
| 77 |
+
if file_name.endswith('.txt'):
|
| 78 |
+
source_path = os.path.join(folder_path, file_name)
|
| 79 |
+
dest_path = os.path.join(folder_path + '/txt/', file_name)
|
| 80 |
+
shutil.move(source_path, dest_path) # Moves txt files
|
| 81 |
+
</code></pre>
|