Mixed/Corrupted Language Transcriptions

[hugoaap-code]

🐞 Bug Report: Mixed/Corrupted Language Transcriptions

🚀 Description

• Despite setting Portuguese as the main language, the transcription frequently displays random or "weird" characters/words from other languages that were not spoken. (Hindi, Russian, english, etc)
• This corruption occurs consistently in nearly every conversation recorded using the L. Pendant with the Omi app.

✅ Expected Behavior

• When a single language (e.g., Portuguese) is explicitly set, the transcription should exclusively use the characters and vocabulary of that language.
• Alternatively, if multi-language transcription is the default, the user should have a clear option to select "Single Language Only" mode to prevent this corruption.

🛑 Actual Behavior

• The transcription output is often garbled with foreign characters or words, even when only the designated language is spoken.

💥 Impact

• Unreadable Transcriptions: The mixed characters/words make the text difficult or impossible to read and use.
• Data Integrity Loss: The transcribed record is corrupted and unreliable.

📝 Environment/Context (Provided by User)

• Device: Samsung S25 Ultra (Android 16, One UI 8.0).
• Setup: Omi app set to never sleep/always run in the background; phone not in battery saving mode; device started, connected, recording initiated, and screen turned off; full battery, not charging.

💡 Suggested Solution

• Provide a user setting to opt for Single/Multi Language transcription (as opposed to a default multi-language mode).

[beastoin]

should be ready in 24h.

[beastoin]

@hugoaap-code Critical STT issue - transcriptions showing random characters from Hindi, Russian, etc. despite language set to Portuguese. This corrupts the data and makes transcripts unusable. A single-language mode option could help. I scored it as Layer understand / P1 / maintainer with score 23 (w=4, sev 4, trust 3, freq 3, lev 2, cost 1). See our triage guide.

---

by AI for @beastoin

[beastoin]

Hi there — this issue has been open for a while with no further updates since the last response. Closing for now to keep things tidy.

If this is still an issue or you'd like to continue the conversation, please feel free to reopen and we'll take another look. Thanks for the report!