Transcribe Audio or Video with Transloadit

Use this for

• Local audio or video files that should become text transcripts or subtitle files
• One or many files, with outputs downloaded locally
• Agent workflows where credentials may already be in the shell, a local .env, or

~/.transloadit/credentials

Inputs

• Absolute path to one or more local audio or video files
• Optional output path; default is the same basename with .txt
• Optional format: text, json, srt, or webvtt

Workflow

1. Confirm each input file exists and resolve it to an absolute path.
2. Derive the output path:

- Single input and no requested output: same directory and basename with .txt - Multiple inputs and no requested output: a transcripts/ directory next to the inputs - Use .json, .srt, or .webvtt when the user asks for that format

1. Let the CLI resolve auth automatically in this order:

- Shell environment variables - The current working directory .env only - ~/.transloadit/credentials If your .env lives in a parent directory, export the variables into the shell first.

1. Run the transcription with the official CLI intent:

npx -y @transloadit/node speech transcribe --input /ABS/PATH/input.opus --output /ABS/PATH/input.txt

For multiple files, repeat --input and write to an output directory:

npx -y @transloadit/node speech transcribe \
  --input /ABS/PATH/a.opus \
  --input /ABS/PATH/b.opus \
  --output /ABS/PATH/transcripts/

Options

• Default format: text
• Other formats: --format json, --format srt, --format webvtt
• Default provider: replicate until ElevenLabs Scribe is ready
• Supported explicit providers: --provider aws, --provider gcp, --provider replicate
• Do not use --provider transloadit; that internal Whisper route is work in progress
• Use --source-language <code> when the input language is known
• Use --target-language <code> when translation is requested

Notes

• Prefer @transloadit/node; it is the official CLI route and exposes speech transcribe.
• If the command errors with a missing subcommand, update to a newer @transloadit/node release

that includes the speech transcription intent.

• Keep secrets local-only; never print TRANSLOADIT_SECRET or copy it into browser code.
• After transcription, confirm each expected output file exists and is non-empty.