mrq/vall-e
5
9
Fork 0
Code Issues 8 Pull Requests Packages Projects Releases Wiki Activity

template cleanup

Browse Source
This commit is contained in:
mrq 2024-12-11 20:06:55 -06:00
parent 7e54e897f7
commit 0c69e798f7
3 changed files with 10 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
data/demo/index.template.html
View File

@ -11,6 +11,7 @@
<tr> <tr>
<th>Text</th> <th>Text</th>
<th>WER↓</th> <th>WER↓</th>
<th>CER↓</th>
<th>SIM-O↑</th> <th>SIM-O↑</th>
<th>Prompt</th> <th>Prompt</th>
<th>Our VALL-E</th> <th>Our VALL-E</th>
@ -27,6 +28,7 @@
<tr> <tr>
<th>Text</th> <th>Text</th>
<th>WER↓</th> <th>WER↓</th>
<th>CER↓</th>
<th>SIM-O↑</th> <th>SIM-O↑</th>
<th>Prompt</th> <th>Prompt</th>
<th>Our VALL-E</th> <th>Our VALL-E</th>

6
vall_e/demo.py
View File

@ -155,9 +155,9 @@ def main():
'Below are some samples from my VALL-E implementation: <a href="https://git.ecker.tech/mrq/vall-e/">https://git.ecker.tech/mrq/vall-e/</a>.', 'Below are some samples from my VALL-E implementation: <a href="https://git.ecker.tech/mrq/vall-e/">https://git.ecker.tech/mrq/vall-e/</a>.',
'Unlike the original VALL-E demo page, I\'m placing emphasis on the input prompt, as the model adheres to it stronger than others.', 'Unlike the original VALL-E demo page, I\'m placing emphasis on the input prompt, as the model adheres to it stronger than others.',
f'Objective metrics are computed by transcribing ({args.transcription_model}) then comparing the word error rate on transcriptions (WER/CER), and computing the cosine similarities on embeddings through a speaker feature extraction model ({args.speaker_similarity_model}) (SIM-O)', f'Objective metrics are computed by transcribing ({args.transcription_model}) then comparing the word error rate on transcriptions (WER/CER), and computing the cosine similarities on embeddings through a speaker feature extraction model ({args.speaker_similarity_model}) (SIM-O)',
'<b>Total WER:</b> ${WER}' '<b>Total WER:</b> ${WER}<br>'
'<b>Total CER:</b> ${CER}' '<b>Total CER:</b> ${CER}<br>'
'<b>Total SIM-O:</b> ${SIM-O}' '<b>Total SIM-O:</b> ${SIM-O}<br>'
]) ])
# comparison kwargs # comparison kwargs

6
vall_e/emb/transcribe.py
View File

@ -9,12 +9,14 @@ import argparse
import torch import torch
import torchaudio import torchaudio
"""
try: try:
import whisperx import whisperx
except Exception as e: except Exception as e:
whisperx = None whisperx = None
print(f"Error while querying for whisperx: {str(e)}") print(f"Error while querying for whisperx: {str(e)}")
pass pass
"""
from transformers import pipeline from transformers import pipeline
@ -193,6 +195,7 @@ def transcribe(
return metadata return metadata
# for backwards compat since it also handles some other things for me # for backwards compat since it also handles some other things for me
"""
def transcribe_whisperx( def transcribe_whisperx(
audio, audio,
language = "auto", language = "auto",
@ -248,6 +251,7 @@ def transcribe_whisperx(
metadata["end"] = end metadata["end"] = end
return metadata return metadata
"""
def transcribe_batch( def transcribe_batch(
input_audio = "voices", input_audio = "voices",
@ -315,7 +319,7 @@ def transcribe_batch(
if os.path.isdir(inpath): if os.path.isdir(inpath):
continue continue
metadata[filename] = transcribe_whisperx( inpath, model_name=model_name, diarize=diarize, device=device, dtype=dtype ) metadata[filename] = transcribe( inpath, model_name=model_name, diarize=diarize, device=device, dtype=dtype )
open(outpath, 'w', encoding='utf-8').write(json.dumps(metadata)) open(outpath, 'w', encoding='utf-8').write(json.dumps(metadata))