Thanks to visit codestin.com
Credit goes to www.scribd.com

Scribd
Upload
13 views4 pages

Audio Transcript Config

The document is a configuration file for audio transcription tasks, detailing various transcription implementations including Vosk, Wav2Vec2, Whisper, Microsoft Azure, and Google services. It includes options for language models, audio conversion commands, and specific settings for each transcription method. Additionally, it provides guidelines for setting up remote services and managing resource requirements for optimal transcription performance.

Uploaded by

Allencar Limma
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as TXT, PDF, TXT or read online on Scribd
13 views4 pages

Audio Transcript Config

The document is a configuration file for audio transcription tasks, detailing various transcription implementations including Vosk, Wav2Vec2, Whisper, Microsoft Azure, and Google services. It includes options for language models, audio conversion commands, and specific settings for each transcription method. Additionally, it provides guidelines for setting up remote services and managing resource requirements for optimal transcription performance.

Uploaded by

Allencar Limma
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as TXT, PDF, TXT or read online on Scribd
You are on page 1/ 4

# Audio transcript task config file.

#########################################
# Choose the transcription implementation
#########################################

# Default implementation uses Vosk transcription on local CPU (slow and medium
quality).
# We include small portable models for 'en' and 'pt-BR', if you want to use a
different language model
# you should download it from https://alphacephei.com/vosk/models and put in
'models/vosk/[lang]' folder.
#implementationClass = iped.engine.task.transcript.VoskTranscriptTask

# Uses a local wav2vec2 implementation for transcription. Accuracy is much better


than most Vosk models.
# This is up to 10x slower than Vosk on high end CPUs. Using a good GPU is highly
recommended!
# Please check the installation steps:
https://github.com/sepinf-inc/IPED/wiki/User-Manual#wav2vec2
# If you enable this, you must set 'huggingFaceModel' param below.
#implementationClass = iped.engine.task.transcript.Wav2Vec2TranscriptTask

# Uses a local Whisper implementation for transcription. Accuracy is better than


wav2vec2 depending on the model.
# This is up to 4x slower than wav2vec2 depending on compared models. Using a high
end GPU is strongly recommended!
# Please check the installation steps:
https://github.com/sepinf-inc/IPED/wiki/User-Manual#whisper
# If you enable this, you must set 'whisperModel' param below.
implementationClass = iped.engine.task.transcript.WhisperTranscriptTask

# Uses a remote service for transcription.


# The remote service is useful if you have a central server/cluster with many GPUs
to be shared among processing nodes.
# Please check steps on https://github.com/sepinf-inc/IPED/wiki/User-Manual#remote-
transcription
# If you enable this, you must set 'remoteServiceAddress' param below.
#implementationClass = iped.engine.task.transcript.RemoteTranscriptionTask

# If you want to use the Microsoft Azure service implementation, comment above and
uncomment below.
# You MUST include Microsoft client-sdk.jar into plugins folder.
# Download it from
https://csspeechstorage.blob.core.windows.net/maven/com/microsoft/
cognitiveservices/speech/client-sdk/1.19.0/client-sdk-1.19.0.jar
# You must pass your subscription key using command line parameter -
XazureSubscriptionKey=XXXXXXXX
#implementationClass = iped.engine.task.transcript.MicrosoftTranscriptTask

# If you want to use the Google Service implementation, comment above and uncomment
below.
# You must include google-cloud-speech-1.22.5.jar AND ITS DEPENDENCIES in plugins
folder.
# You can download an all-in-one jar from https://gitlab.com/iped-project/iped-
maven/-/blob/master/com/google/cloud/google-cloud-speech/1.22.5-shaded/google-
cloud-speech-1.22.5-shaded.jar
# Finally you must set environment variable GOOGLE_APPLICATION_CREDENTIALS pointing
to your credential file
#implementationClass = iped.engine.task.transcript.GoogleTranscriptTask

#########################################
# Global options
#########################################

# Language model(s) to use when processing audios. 'auto' uses the 'locale' set on
LocalConfig.txt
# You can specify one or two languages separated by ; if Microsoft or Google are
used.
# Vosk implementation accepts just one language for now.
# Wav2Vec2 local or remote implementation don't use this, you must set
'huggingFaceModel' below
# Setting more than 1 lang model can result in wrong language detection.
language = detect

# Command to convert audios to wav before transcription. Do not change $INPUT or


$OUTPUT params.
convertCommand = mplayer -benchmark -vo null -vc null -srate 16000 -af
format=s16le,resample=16000,channels=1 -ao pcm:fast:file=$OUTPUT $INPUT

# Mime types or supertypes to process. If you want to add videos use ; as separator
and update 'convertCommand'.
mimesToProcess = audio/3gpp; audio/3gpp2; audio/vnd.3gpp.iufp; audio/x-aac;
audio/x-aiff; audio/amr; audio/amr-wb; audio/amr-wb+; audio/mp4; audio/ogg;
audio/vorbis; audio/x-oggflac; audio/x-oggpcm; audio/opus; audio/speex;
audio/qcelp; audio/vnd.wave; audio/x-caf; audio/x-ms-wma; audio/x-opus+ogg;
audio/ilbc

# Skip known audios found in the hash lookup database.


skipKnownFiles = true

# Minimum number of seconds to wait for each audio transcription.


minTimeout = 180

# Number of seconds to wait for each audio second transcription. 'minTimeout' param
above is added to this.
timeoutPerSec = 3

#########################################
# VoskTranscriptTask options
#########################################

# Minimum word score to include the word in transcription result. Applies just to
Vosk implementation.
# If you don't want to see * in transcription results, set this to 0
minWordScore = 0.5

#########################################
# Local Wav2Vec2TranscriptTask options
#########################################

# HuggingFace model card to be used by Wav2Vec2TranscriptTask. You must uncomment


one of them.

# Small models for portuguese, ~23-24% WER on tested data sets


# huggingFaceModel = lgris/bp_400h_xlsr2_300M
# huggingFaceModel = Edresson/wav2vec2-large-xlsr-coraa-portuguese
# Small models for other languages, WER not evaluated
# huggingFaceModel = jonatasgrosman/wav2vec2-large-xlsr-53-english
# huggingFaceModel = jonatasgrosman/wav2vec2-large-xlsr-53-german
# huggingFaceModel = jonatasgrosman/wav2vec2-large-xlsr-53-italian
# huggingFaceModel = jonatasgrosman/wav2vec2-large-xlsr-53-spanish
# huggingFaceModel = jonatasgrosman/wav2vec2-large-xlsr-53-french

# Bigger models, better on tested pt-BR data sets, but uses more RAM and is ~2x
slower.
# The portuguese model has ~19% WER on tested data sets.
# huggingFaceModel = jonatasgrosman/wav2vec2-xls-r-1b-portuguese
# huggingFaceModel = jonatasgrosman/wav2vec2-xls-r-1b-english
# huggingFaceModel = jonatasgrosman/wav2vec2-xls-r-1b-german
# huggingFaceModel = jonatasgrosman/wav2vec2-xls-r-1b-italian
# huggingFaceModel = jonatasgrosman/wav2vec2-xls-r-1b-spanish
# huggingFaceModel = jonatasgrosman/wav2vec2-xls-r-1b-french

#########################################
# Local WhisperTranscriptTask options
#########################################

# Possible values: tiny, base, small, medium, large-v3, dwhoelz/whisper-large-pt-


cv11-ct2
# large-v3 is much better than medium, but 2x slower and uses 2x more memory.
# If you know the language you want to transcribe, please set the 'language' option
above.
# 'language = auto' uses the 'locale' set on LocalConfig.txt
# 'language = detect' uses auto detection, but it can cause mistakes
whisperModel = medium

# Which device to use: 'cpu' or 'gpu'. For 'gpu', please see the official faster-
whisper documentation
# for the installation requirements: https://github.com/SYSTRAN/faster-whisper?
tab=readme-ov-file#gpu
device = cpu

# Compute type precision. This affects accuracy, speed and memory usage.
# Possible values: float32 (better), float16 (recommended for GPU), int8 (faster)
precision = int8

# Batch size (number of parallel transcriptions). If you have a GPU with enough
memory,
# increasing this value to e.g. 16 can speed up transcribing long audios up to 10x.
# Test what is the better value for your GPU before hitting OOM.
# This works just if you are using whisperx library instead of faster_whisper
batchSize = 1

#########################################
# RemoteAudioTranscriptTask options
#########################################

# IP:PORT of the service/central node used by the RemoteWav2Vec2TranscriptTask


implementation.
# remoteServiceAddress = 127.0.0.1:11111

#########################################
# MicrosoftTranscriptTask options
#########################################
# Specific for Microsoft service. Replace with your own subscription region
identifier from here: https://aka.ms/speech/sdkregion
serviceRegion = brazilsouth

# Depending on your subscription, Microsoft limits the number of max concurrent


requests
maxConcurrentRequests = 100

#########################################
# GoogleTranscriptTask options
#########################################

# Depending on your subscription, Google limits your request rate (per minute or
per second).
requestIntervalMillis = 67

# Set the Google transcription model to be used.


# Some possible values: default, phone_call, video, latest_short, latest_long
# For more information, see
https://cloud.google.com/speech-to-text/docs/transcription-model
googleModel = latest_long

You might also like