Elevenlabs Transcribe

Transcribe audio to text using ElevenLabs Scribe. Supports batch transcription, realtime streaming from URLs, microphone input, and local files.

Security Scan

The skill's code and requirements match the claimed transcription functionality, but the package mislabels itself as 'Official ElevenLabs' while the source/publisher is unknown and there are a few minor supply-chain/credential-read concerns you should review before installing.

Purpose & Capability

The scripts and declared requirements (ffmpeg, python3, ELEVENLABS_API_KEY) align with a speech-to-text skill using ElevenLabs. However, SKILL.md calls this the 'Official ElevenLabs skill' while the registry 'Source' is unknown and the owner ID does not obviously belong to ElevenLabs — possible impersonation or mislabeling.

Instruction Scope

The runtime instructions and scripts stay within the stated purpose: convert audio (file, mic, URL) to text and send audio to ElevenLabs via their SDK. One minor scope note: the Python code calls load_dotenv(), which will read a local .env file if present — that can surface other environment variables from disk (not declared in requires.env).

Install Mechanism

There is no platform install spec, but the provided shell wrapper auto-creates a local virtualenv and runs pip install -r requirements.txt. Main dependencies are pinned with hashes for supply-chain integrity (elevenlabs, pydub, python-dotenv), but some platform-specific packages (sounddevice, numpy) are not hashed. pip installs from PyPI on first run (network activity) and writes a .venv directory under the skill folder.

Credentials

Only ELEVENLABS_API_KEY is declared and used; that is appropriate for a transcription client. Note that load_dotenv() may read a .env file from disk and load additional env vars implicitly. The code will transmit audio and the API key (via the ElevenLabs SDK) to ElevenLabs' service — this is expected behavior but worth confirming you're comfortable sending audio to that provider.

Persistence & Privilege

The skill does not request always:true and won't be force-included. It sets up a per-skill .venv and an installed marker in the skill directory; it doesn't modify other skills or system-wide agent settings.

What to consider before installing

This skill's code behaves like a normal ElevenLabs transcription client, but before installing: 1) Confirm the publisher — SKILL.md claims 'Official ElevenLabs' but the source/owner are not ElevenLabs; prefer official plugins from the vendor when possible. 2) Review and protect your ELEVENLABS_API_KEY (use a scoped/test key if possible). 3) Be aware the script will create a local .venv and pip-install packages from PyPI (network activity); consider installing in an isolated environment/container. 4) Note load_dotenv() will read a .env file in the skill directory and could load other env vars — remove secrets you don't want read. 5) If you need stronger supply-chain guarantees, request that all requirements be pinned with verified hashes for every platform-specific package or run the code review/install inside an isolated sandbox first.

ElevenLabs Speech-to-Text

Official ElevenLabs skill for speech-to-text transcription.

Convert audio to text with state-of-the-art accuracy. Supports 90+ languages, speaker diarization, and realtime streaming.

Prerequisites

• ffmpeg installed (brew install ffmpeg on macOS)
• ELEVENLABS_API_KEY environment variable set
• Python 3.8+ (dependencies auto-install on first run)

Usage

{baseDir}/scripts/transcribe.sh <audio_file> [options]
{baseDir}/scripts/transcribe.sh --url <stream_url> [options]
{baseDir}/scripts/transcribe.sh --mic [options]

Examples

Batch Transcription

Transcribe a local audio file:

{baseDir}/scripts/transcribe.sh recording.mp3

With speaker identification:

{baseDir}/scripts/transcribe.sh meeting.mp3 --diarize

Get full JSON response with timestamps:

{baseDir}/scripts/transcribe.sh interview.wav --diarize --json

Realtime Streaming

Stream from a URL (e.g., live radio, podcast):

{baseDir}/scripts/transcribe.sh --url https://npr-ice.streamguys1.com/live.mp3

Transcribe from microphone:

{baseDir}/scripts/transcribe.sh --mic

Stream a local file in realtime (useful for testing):

{baseDir}/scripts/transcribe.sh audio.mp3 --realtime

Quiet Mode for Agents

Suppress status messages on stderr:

{baseDir}/scripts/transcribe.sh --mic --quiet

Options

Option	Description
--diarize	Identify different speakers in the audio
--lang CODE	ISO language hint (e.g., en, pt, es, fr)
--json	Output full JSON with timestamps and metadata
--events	Tag audio events (laughter, music, applause)
--realtime	Stream local file instead of batch processing
--partials	Show interim transcripts during realtime mode
-q, --quiet	Suppress status messages (recommended for agents)

Output Format

Text Mode (default)

Plain text transcription:

The quick brown fox jumps over the lazy dog.

JSON Mode (--json)

{
  "text": "The quick brown fox jumps over the lazy dog.",
  "language_code": "eng",
  "language_probability": 0.98,
  "words": [
    {"text": "The", "start": 0.0, "end": 0.15, "type": "word", "speaker_id": "speaker_0"}
  ]
}

Realtime Mode

Final transcripts print as they're committed. With --partials:

[partial] The quick
[partial] The quick brown fox
The quick brown fox jumps over the lazy dog.

Supported Formats

Audio: MP3, WAV, M4A, FLAC, OGG, WebM, AAC, AIFF, Opus Video: MP4, AVI, MKV, MOV, WMV, FLV, WebM, MPEG, 3GPP

Limits: Up to 3GB file size, 10 hours duration

Error Handling

The script exits with non-zero status on errors:

• Missing API key: Set ELEVENLABS_API_KEY environment variable
• File not found: Check the file path exists
• Missing ffmpeg: Install with your package manager
• API errors: Check API key validity and rate limits

When to Use Each Mode

Scenario	Command
Transcribe a recording	./transcribe.sh file.mp3
Meeting with multiple speakers	./transcribe.sh meeting.mp3 --diarize
Live radio/podcast stream	./transcribe.sh --url <url>
Voice input from user	./transcribe.sh --mic --quiet
Need word timestamps	./transcribe.sh file.mp3 --json