Connector library

Implemented source Voice, SMS & Messaging

Local Speech Recognition connector.

Submit and retrieve offline recording transcription jobs through the CXFabric HTTPS controller using a configured local faster-whisper model.

Connector download

Download Local Speech Recognition

Versioned ZIP for your CXFabric tenant.

Requires a separately deployed controller. Download controller source and setup instructions after signing in. Import only the connector ZIP into your tenant.

Use this connector in a flow

  1. Register the ZIP in your tenant, or add a new version if the connector is already installed.
  2. Create or select a credential if the action requires one.
  3. In a Draft flow, add an available action, choose its version, connect it, and map inputs.
  4. Test the step and flow; review output and failures before activation.

See readme.md in the ZIP for settings and limitations.

Version
0.1.0.0
Release
2026-10-17
Package
Implemented source
Schema credentials
Bearer token
Connection settings
Controller URL
Package size
11 KB
SHA-256
34914e3d0db92925aa3c31c6031b9db446e8c7fd8e67cd5650df40c64d9b55cc
Notes
Read release notes

Included in this version

Included actions

Implemented in v0.1.0.0. Use a separate flow trigger; no webhook trigger is included.

Submit Transcription

Start one local, offline recording transcription job. Retrieve its status in a later action; submission is not completion.

Get Transcription

Read an owned transcription job and its completed text, timestamps and channel qualification. Failed processing returns an error.

Vendor API

API and access

Implementation considerations
  • Submit Transcription: Start one local, offline recording transcription job. Retrieve its status in a later action; submission is not completion.
  • Get Transcription: Read an owned transcription job and its completed text, timestamps and channel qualification. Failed processing returns an error.
  • Deploy CXFabric FreeSWITCH Controller 0.2.0 before using these APIs. Download its separate source archive and setup/lab guide from this connector’s download page. No hosted endpoint, carrier account, infrastructure or model weights are provided.
  • Use a controller-issued Bearer Token credential and trusted Controller URL from Flow Properties. Use the same credential across related call, media, transcription and evaluation actions. Requests are bounded to 1–8 seconds; asynchronous processing is retrieved in later steps without internal polling or retries.
  • One CPU job runs at a time with an operator-configured deadline. Retain Request IDs to reconcile identical submissions. Restarted jobs fail rather than resubmit. Inspect status; a completed transcript is not a passing assertion. Select a channel explicitly; qualified IVR attribution requires an actual stereo-channel lab check.
  • The optional local recognition worker uses a separately installed pinned faster-whisper package and a staged local model. The base controller image does not include the Python/ML runtime. Actual model inference, the optional speech image and resource capacity remain unvalidated.
  • Local package and native HTTPS/SQLite integration checks use synthetic switch/model evidence. Hosted CXFabric import, live FreeSWITCH/carrier audio, recording-channel attribution and recognition accuracy have not been qualified.

Support

Need help with Local Speech Recognition?

Tell us what you need to connect or change.