Speech-to-Text system

A key service in the building of interactive voice systems.
Now, there are many online services with similar functionality, but most of them bring the risks of data leaks, randomly latencies on request processing, and requires a paid subscription.
This software is free from the enumerated shortcomings and enables to build complete local speech-to-text service.
This is a fast, lightweight solution with simple web-api which doesn't require some special hardware (such as GPU) and manage to operate on regular servers with appropriate performance.

Written in pure C, with libs: mpg123, alphacephei framework.

---

Price: 150$ / 150 USDT
For purchase questions, please visit contact page.
A trial period with installation on your servers is provided (preferred Ubuntu 22.04 x64).

---

Basic features:

• Complete local system

  You won't depend on any online services, and spend money for subscription

• Supports multiple nanguages

  There are open models for various languages

• Supports models fine-tuning

  There are tools for it

• Context dictionary

  Allows to defined the context dictionary (bag of accessible words) an use it during recognition

• Speaker identify

  Allows to generate speakers signature vector for further identification

• Capable to run on regular servers

  IBM x3550-M3 or event Core2 Quad with 2Gb RAM (like my local PBX for example)

• Supported in FreeSWITCH

  There is already a ready-made module for integration with FreeSWITCH.
  for details see: mod_sivr_stt

• Models preloading and caching

  Takes a positive effect on performance and memory consumption

• Simple web api

  Enable to make easy integration with various applications

• Supported formats

  - wav
  - mp3
  - l16

• Supported os

  - Linux

--- Examples ---

Example #1 (simple request)

Request:
curl http://127.0.0.1:8801/v1/transcriptions -X POST -H "Authorization: Bearer secret" -H "Content-Type: multipart/form-data" -F language="en" -F smodel="small" -F file="@test.mp3"

Response (json):
{
 "text" : "hello world"
 }
        

Example #2 (with speake identify)

Request:
curl http://127.0.0.1:8801/v1/transcriptions -X POST -H "Authorization: Bearer secret" -H "Content-Type: multipart/form-data" -F language="en" -F smodel="small" -F vmodel="default" -F file="@test.mp3"

Response (json):
{
 "spk" : [-0.644623, 1.023342, 2.575434, 0.623447, -0.602342, 1.0234234 -1.4824234 -0.021242, 0.824297, -0.152424, ... ],
 "spk_frames" : 81,
 "text" : "hello world"
}