kitlangton / kitlangton/Hex

Bug report: Multilingual Parakeet TDT v3 model mixes Languages with English

Open
#152 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Swift
Stars
2.9k
Forks
226
PR merge metrics
No merged PRs in 30d

Description

**Is your bug related to a problem? Please describe.**
When transcribing audio entirely in Spanish using Hex version 0.6.9 and the provided multilingual Parakeet TDT v3 model, the output incorrectly mixes Spanish and English elements. This behavior appears to be specific to the implementation in Hex, as the same model performs correctly in other applications like Spokenly.

**Describe the solution you'd like**
The transcription should accurately reflect the source language (Spanish) without injecting English words or phrases when the input is strictly in Spanish.

**Describe alternatives you've considered**
I have tested the same model in Spokenly, where the transcription is handled correctly without language mixing.

**Additional context**
* **Application Version:** Hex 0.6.9
* **Model:** Multilingual Parakeet TDT v3
* **Observed Behavior:** Language "hallucination" or mixing (Spanish/English) in mono-lingual Spanish tasks.

Ejemplo:
El primer destacamiento operacional de Fuerzas Delta, denominado indistintamente Fuerza Delta, Grupo de Aplicaciones de Combate, es una fuerza de operaciones especiales del ejército de Estados United bajo control operacional and commando conjunto oferaciones especiales. The missionary principal of the Fuerza Delta are electronterrorismo, rescate de rehenes, acciones directs and reconocimiento special, a menudo contra objetos de alta priority.

Contributor guide

No contributing guide indexed for this repository

Research direction

No source file, test, or entry point is named. Start by locating the legacy Swift implementation used by Hex 0.6.9 for the multilingual Parakeet TDT v3 model, then reproduce the Spanish-only example and compare its output with Spokenly; done means Spanish audio no longer produces English or mixed-language text.

Written by the indexing model from the issue text.

Assessment

Tech stack
macos, swift
Domain
audio-video-rtc, desktop
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.