Transcribe
2 minute read
LocalStack Community ships with an offline speech-to-text service powered by Vosk.
LocalStack requires an internet connection the first time a transcription job is created for a given language. This is to download and cache the language model. Subsequent transcriptions for the same language can be done offline. Language models are around 50 MiB each and saved to the cache directory (see Filesystem Layout).
Note
This service has limited support for aarch64/Apple Silicon.
If you encounter cannot load library *.so errors, please try the AMD64 build of LocalStack:
$ docker pull localstack/localstack:2.0.0 --platform amd64Supported Formats
The following input media formats are supported:
- Adaptive Multi-Rate (AMR)
- Free Lossless Audio Codec (FLAC)
- MPEG-1 Audio Layer-3 (MP3)
- MPEG-4 Part 14 (MP4)
- OGG
- Matroska Video files (MKV)
- Waveform Audio File Format (WAV)
Supported Languages
The following langauges and dialects are supported:
| Language | Language Code |
|---|---|
| German | de-DE |
| English, British | en-GB |
| English, Indian | en-IN |
| English, US | en-US |
| Spanish | es-ES |
| Farsi | fa-IR |
| French | fr-FR |
| Hindi | hi-IN |
| Italian | it-IT |
| Japan | ja-JP |
| Dutch | nl-NL |
| Portuguese | pt-BR |
| Russian | ru-RU |
| Turkish | tr-TR |
| Vietnamese | vi-VN |
| Chinese | zh-CN |
Transcribing Audio
Create an S3 bucket and upload the audio file:
$ awslocal s3 mb s3://foo
$ awslocal s3 cp ~/example.wav s3://foo/example.wavCreate the transcription job:
$ awslocal transcribe start-transcription-job \
--transcription-job-name example \
--media MediaFileUri=s3://foo/example.wav \
--language-code en-INJobs can be listed like so:
$ awslocal transcribe list-transcription-jobs
{
"TranscriptionJobSummaries": [
{
"TranscriptionJobName": "example",
"CreationTime": "2022-08-17T14:04:39.277000+05:30",
"StartTime": "2022-08-17T14:04:39.308000+05:30",
"LanguageCode": "en-IN",
"TranscriptionJobStatus": "IN_PROGRESS"
}
]
}Once job is complete, the transcript can be retrieved from the S3 bucket:
$ awslocal transcribe get-transcription-job --transcription-job example
{
"TranscriptionJob": {
"TranscriptionJobName": "example",
"TranscriptionJobStatus": "COMPLETED",
"LanguageCode": "en-IN",
"MediaFormat": "wav",
"Media": {
"MediaFileUri": "s3://foo/example.wav"
},
"Transcript": {
"TranscriptFileUri": "s3://foo/7844aaa5.json"
},
"CreationTime": "2022-08-17T14:04:39.277000+05:30",
"StartTime": "2022-08-17T14:04:39.308000+05:30",
"CompletionTime": "2022-08-17T14:04:57.400000+05:30",
}
}
$ awslocal s3 cp s3://foo/7844aaa5.json .
$ jq .results.transcripts[0].transcript 7844aaa5.json
"it is just a question of getting rid of the illusion that we are separate from nature"