SDK-first: always use the official SDK — see gladia-sdk-integration for policy, setup, and fallback criteria.

References

Consult these sibling skills as needed:

• ../gladia-sdk-integration/SKILL.md -- SDK setup, client initialization, error handling, and SDK vs raw API decision guide
• ../gladia-sdk-integration/references/sdk-versions.md -- Current SDK versions (auto-synced by CI)
• ../gladia-troubleshooting/SKILL.md -- Common errors, gotchas, and verification checklist
• ../gladia-live-transcription/SKILL.md -- Live streaming transcription
• ../gladia-pre-recorded-transcription/SKILL.md -- Pre-recorded file transcription

---

name: Gladia description: Use when building speech-to-text transcription features, processing audio or video files, implementing real-time transcription, extracting insights from audio (speaker identification, translation, sentiment, summaries), or integrating transcription into applications via API or SDK. metadata: mintlify-proj: gladia version: "1.0"

Gladia Skill

Product summary

Gladia is a speech-to-text (STT) API that transcribes audio and video files in two modes: pre-recorded (asynchronous, for files) and live (real-time, for streaming audio). Beyond transcription, it offers audio intelligence features like speaker diarization, translation, sentiment analysis, summarization, and named entity recognition. Use the JavaScript/TypeScript SDK (@gladiaio/sdk) or Python SDK (gladiaio-sdk) for simplified integration, or call the REST API directly. Authenticate with the x-gladia-key header. Primary docs: https://docs.gladia.io

Key endpoints:

• Pre-recorded: POST /v2/pre-recorded (create job), GET /v2/pre-recorded/:id (get result)
• Live: POST /v2/live (init session), WebSocket connection for streaming
• Upload: POST /v2/upload (for local files)

Models: solaria-3 (pre-recorded, highest quality), solaria-1 (live only)

When to use

Reach for Gladia when:

• Transcribing files: User uploads an audio/video file and needs a transcript (meetings, podcasts, interviews, call recordings)
• Real-time transcription: Streaming audio from a microphone, phone call, or live event that needs instant captions or transcripts
• Extracting insights: Need to identify speakers, translate to other languages, detect sentiment, summarize, or extract entities from audio
• Multi-language support: Audio contains multiple languages or needs translation to 100+ target languages
• Subtitle generation: Creating SRT/VTT subtitle files for video content
• Integration: Building a feature into an app or workflow (via SDK or API)

Do not use Gladia for: text-only processing, image analysis, or non-audio content.

Quick reference

Authentication

# All requests require the x-gladia-key header
curl -H "x-gladia-key: YOUR_API_KEY" https://api.gladia.io/v2/...

Pre-recorded workflow (SDK)

const { GladiaClient } = require("@gladiaio/sdk");
const client = new GladiaClient({ apiKey: "YOUR_KEY" });

// One-call transcription
const result = await client.preRecorded().transcribe("audio.mp3", {
  model: "solaria-3",
  language_config: { languages: ["en"] },
  diarization: true,
  translation: true,
  translation_config: { target_languages: ["fr"] }
});

Live workflow (SDK)

const config = {
  model: "solaria-1",
  encoding: "wav/pcm",
  sample_rate: 16000,
  bit_depth: 16,
  channels: 1,
  language_config: { languages: ["en"] }
};

const session = client.liveV2().startSession(config);
session.on("message", (msg) => console.log(msg));
session.sendAudio(audioChunk);
session.stopRecording();

Supported audio formats

MP3, WAV, FLAC, OGG, OPUS, AAC, M4A, AC3, EAC3, MP2

Supported video formats

MP4, MOV, AVI, MKV, FLV, 3GP, WMV, M4V

Language codes

Use 2-letter ISO 639-1 codes: en, fr, de, es, zh, ja, etc. Full list at /chapters/language/supported-languages

Audio intelligence features

Feature	Pre-recorded	Live	Config key
Speaker diarization	✓	✗	diarization
Translation	✓	✓	translation
Sentiment analysis	✓	✗	sentiment_analysis
Summarization	✓	✗	summarization
Named entity recognition	✓	✓	named_entity_recognition
PII redaction	✓	✗	pii_redaction
Subtitles (SRT/VTT)	✓	✗	subtitles
Chapterization	✓	✗	chapterization
Custom vocabulary	✓	✓	custom_vocabulary

Decision guidance

Pre-recorded vs. Live

Scenario	Use pre-recorded	Use live
User uploads a file	✓	✗
Streaming microphone input	✗	✓
Phone call transcription	✗	✓
Podcast/interview processing	✓	✗
Real-time captions	✗	✓
Need diarization	✓	✗ (use multi-channel instead)
Need sentiment/summarization	✓	✗

Model selection

Model	Use case	Availability
solaria-3	Highest accuracy, best for meetings/calls/podcasts	Pre-recorded only
solaria-1	Real-time streaming, lower latency	Live only

Language detection vs. explicit language

Approach	When to use
Explicit language_config.languages: ["en"]	Language is known; avoids detection overhead
Auto-detect (omit languages)	Language unknown or mixed; slower but flexible
Code switching code_switching: true	Audio mixes multiple languages; requires language list

Diarization vs. multi-channel

Approach	When to use
diarization: true	Single audio stream, multiple speakers (meetings, interviews)
Multi-channel (channels > 1)	Separate audio tracks per speaker (phone calls with distinct channels)

Workflow

Pre-recorded transcription (file upload)

1. Prepare the file: Ensure audio/video is in a supported format (MP3, WAV, MP4, etc.) and under 1000 MB. For files >135 minutes, split into ~60-minute chunks.

2. Upload the file (if local):

   curl -X POST https://api.gladia.io/v2/upload \
     -H "x-gladia-key: YOUR_KEY" \
     -F "audio=@audio.mp3"
   

   Save the returned audio_url.

3. Create transcription job:

   curl -X POST https://api.gladia.io/v2/pre-recorded \
     -H "x-gladia-key: YOUR_KEY" \
     -H "Content-Type: application/json" \
     -d '{
       "audio_url": "https://api.gladia.io/file/...",
       "model": "solaria-3",
       "language_config": { "languages": ["en"] },
       "diarization": true
     }'
   

   Save the returned id.

4. Poll for results:

   curl https://api.gladia.io/v2/pre-recorded/ID \
     -H "x-gladia-key: YOUR_KEY"
   

   Repeat until status: "done". Or use webhooks/callbacks instead of polling.

5. Extract data: Parse result.transcription.utterances for text, timing, speaker, language, and confidence.

Live transcription (streaming)

1. Initialize session:

   curl -X POST https://api.gladia.io/v2/live \
     -H "x-gladia-key: YOUR_KEY" \
     -H "Content-Type: application/json" \
     -d '{
       "encoding": "wav/pcm",
       "sample_rate": 16000,
       "bit_depth": 16,
       "channels": 1
     }'
   

   Save the returned WebSocket url.

2. Connect WebSocket: Open a WebSocket connection to the URL.

3. Send audio chunks: Stream audio in the specified encoding/sample rate.

4. Read messages: Listen for transcript, translation, sentiment_analysis, etc. messages.

5. Stop recording: Send { "type": "stop_recording" } or close the socket with code 1000.

6. Retrieve final results: Call GET /v2/live/:id to get the complete result.

Common gotchas

• Model mismatch: solaria-3 is pre-recorded only; solaria-1 is live only. Using the wrong model will fail.
• Language config with solaria-3: Must set exactly one language in language_config.languages (e.g., ["fr"]). Do not pass multiple languages or enable code switching with solaria-3.
• Polling timeout: Pre-recorded jobs can take minutes. Don't assume instant results; use webhooks or callbacks for production.
• File size limits: Max 1000 MB per file; max 135 minutes per job (enterprise plans support 4h15). Split large files.
• Live session duration: Max 3 hours per WebSocket session. Start a new session before hitting the limit.
• Audio format mismatch: Ensure encoding, sample_rate, bit_depth, and channels match your actual audio stream, or transcription will fail silently.
• Diarization hints are not constraints: number_of_speakers, min_speakers, max_speakers are hints, not hard limits. The model may detect a different count.
• Translation with code switching: Do not enable both code_switching: true and translation on the same request without a constrained language list.
• Concurrency limits: Free tier: 3 concurrent pre-recorded, 1 live. Paid: 25 concurrent pre-recorded, 30 live. Queued requests beyond concurrency limits will wait.
• Deprecated V1 API: Old endpoints like /audio/text/audio-transcription/ no longer work. Use /v2/pre-recorded and /v2/live.

Verification checklist

Before submitting transcription work:

• API key is valid and passed in x-gladia-key header
• Audio file is in a supported format (MP3, WAV, MP4, etc.)
• File size is under 1000 MB; duration under 135 minutes (or split into chunks)
• For pre-recorded: model is solaria-3; for live: model is solaria-1
• Language code is valid ISO 639-1 (e.g., en, fr, de)
• For solaria-3: exactly one language in language_config.languages
• Audio encoding, sample rate, bit depth, and channels match the actual stream (live only)
• Diarization is enabled if speaker identification is needed
• Translation target languages are in the supported list
• Webhook/callback URL is reachable (if using async notifications)
• Job status is done before reading results
• Response contains result.transcription.utterances with text and timing

Resources

• Full page navigation: https://docs.gladia.io/llms.txt
• API Reference: https://docs.gladia.io/api-reference/
• Pre-recorded quickstart: https://docs.gladia.io/chapters/pre-recorded-stt/quickstart
• Live transcription quickstart: https://docs.gladia.io/chapters/live-stt/quickstart
• Audio intelligence features: https://docs.gladia.io/chapters/audio-intelligence/
• Recommended parameters by use case: https://docs.gladia.io/chapters/pre-recorded-stt/recommended-parameters
• Supported languages: https://docs.gladia.io/chapters/language/supported-languages
• SDK (JavaScript): https://www.npmjs.com/package/@gladiaio/sdk
• SDK (Python): https://pypi.org/project/gladiaio-sdk/

---

For additional documentation and navigation, see: https://docs.gladia.io/llms.txt

---

This file is auto-synced from https://docs.gladia.io/.well-known/agent-skills/gladia/skill.md Do not edit manually — changes will be overwritten by CI. For additional documentation and navigation, see: https://docs.gladia.io/llms.txt