Skribby
GuidesUpdated 2 hours ago

Transcription Models

Skribby uses a unified naming convention for all supported transcription models to make it easier to understand which provider and model you are using. This guide explains how to format your transcription model keys, understand aliases, and what to expect in API responses.

Standardized Naming Convention

Skribby uses the format <provider>/<model> for all official transcription model keys.

When specifying a transcription_model in the API, we highly recommend using the official key for all new integrations:

{ "transcription_model": "deepgram/nova-2-realtime", "meeting_url": "...", "service": "gmeet", "bot_name": "My Meeting Bot" }

Supported Models & Aliases

We maintain backwards compatibility with older, simpler naming conventions (like whisper or deepgram). These are called aliases. When you request an alias, Skribby internally maps it to the corresponding official key.

Below is the full mapping of official keys to their supported aliases:

Official KeySupported AliasesType
nonenoneNo Transcription
openai/whisper-large-v3whisperStandard
soniox/stt-async-v4soniox, soniox/latestStandard
soniox/stt-rt-v4soniox-realtime, soniox/latest-realtimeRealtime
assemblyai/universal-2assembly-aiStandard
assemblyai/universal-streamingassembly-ai-realtimeRealtime
deepgram/nova-2deepgramStandard
deepgram/nova-2-realtimedeepgram-realtimeRealtime
deepgram/nova-3-multilingualdeepgram-v3Standard
deepgram/nova-3-multilingual-realtimedeepgram-realtime-v3Realtime
speechmatics/batch-standardspeechmaticsStandard
speechmatics/realtime-standardspeechmatics-realtimeRealtime
revai/reverb-foreign-languagerev-aiStandard
elevenlabs/scribe-v1elevenlabsStandard
elevenlabs/scribe-v2elevenlabs-v2Standard
elevenlabs/scribe-v2-realtimeelevenlabs-realtime-v2Realtime
gladia/asyncgladiaStandard
gladia/realtimegladia-realtimeRealtime
salad/transcriptionsaladStandard

API Responses

When you create a bot or recording, the Skribby API response includes two fields relating to your model selection:

  1. transcription_model: This field echoes the exact string you provided in the request. If you sent an alias, this field will show the alias.
  2. transcription_model_key: This field always returns the canonical, official key that Skribby actually used to run the transcription.

This design ensures that your application won't break if it expects transcription_model to match your request, while still giving you visibility into the exact provider and model version being used under the hood.

Example 1: Request with Official Key

Request

{ "transcription_model": "deepgram/nova-2" // ... }

Response

{ "transcription_model": "deepgram/nova-2", "transcription_model_key": "deepgram/nova-2" // ... }

Example 2: Request with Legacy Alias

Request

{ "transcription_model": "whisper" // ... }

Response

{ "transcription_model": "whisper", "transcription_model_key": "openai/whisper-large-v3" // ... }

Example 3: Request with Semantic Alias

Some aliases, like soniox/latest, are designed to always point to a provider's best or most recent model.

Request

{ "transcription_model": "soniox/latest" // ... }

Response

{ "transcription_model": "soniox/latest", "transcription_model_key": "soniox/stt-async-v4" // ... }