Is your feature request related to a problem? Please describe.
The primary motivation for this feature is to help users who attend meetings in a language that is not their native language.
Following a fast technical conversation in a second language can be difficult, especially when participants have different accents, speak quickly, interrupt each other, or use unfamiliar terminology. Real-time captions would allow users to read what is being said while listening, making meetings easier to understand and reducing the risk of missing important information.
NoJoin currently focuses on processing recordings after they have been completed or imported. Although the final transcript is useful, it does not help the user understand the conversation while the meeting is taking place.
Real-time transcription would also support continuous dictation, accessibility workflows, noisy environments, and early detection of transcription errors.
Describe the solution you'd like
Add a Live Mode to the NoJoin interface for real-time transcription.
The interface should display the transcript continuously while audio is being captured. Ideally, it could support two display modes:
- Sentence-by-sentence transcription, showing completed and stable sentences.
- Low-latency partial transcription, updating the text word by word or phrase by phrase, similar to the Live Captions feature in Windows 11.
A possible workflow would be:
- Select the microphone, system audio, or combined audio source.
- Select the transcription language and model.
- Optionally select a target translation language.
- Start a Live Mode session.
- Display partial transcription results immediately.
- Replace partial text with stable sentences as more context becomes available.
- Save the final audio and corrected transcript as a regular NoJoin recording when the session ends.
Useful options could include:
- Adjustable latency versus accuracy.
- Automatic scrolling.
- Large-text or full-screen caption view.
- Displaying the original transcription and live translation side by side.
- Copying the current transcript while the session is active.
- Timestamps for finalized sentences.
- Clear visual distinction between provisional and finalized text.
- Optional speaker labels when real-time diarization is available.
The feature should use local inference when NoJoin is configured for local processing.
Describe alternatives you've considered
Windows 11 Live Captions can provide real-time captions, but its transcript is not integrated with NoJoin and cannot easily become part of the same meeting, notes, speaker, and search workflow.
External captioning, dictation, and translation applications were also considered, but they create a separate workflow and require users to manually import or combine their results afterward.
Processing short audio chunks as separate recordings is another possibility, but it produces fragmented transcripts and is not suitable for a continuous meeting.
Additional context
The main use case is helping people follow meetings conducted in a language that is not their native language.
Other relevant use cases include:
- Live meeting captions.
- Optional real-time translation.
- Continuous personal dictation.
- Accessibility and hearing support.
- Monitoring transcription quality during recording.
- Following conversations in noisy environments.
- Producing a searchable transcript without waiting for post-processing.
The first implementation could prioritize sentence-by-sentence transcription. Word-by-word partial results could be added as an optional lower-latency mode if the selected transcription backend supports streaming or incremental decoding.
When the session ends, NoJoin should perform a final transcription pass if necessary so that the stored transcript prioritizes accuracy over the provisional live output.
Is your feature request related to a problem? Please describe.
The primary motivation for this feature is to help users who attend meetings in a language that is not their native language.
Following a fast technical conversation in a second language can be difficult, especially when participants have different accents, speak quickly, interrupt each other, or use unfamiliar terminology. Real-time captions would allow users to read what is being said while listening, making meetings easier to understand and reducing the risk of missing important information.
NoJoin currently focuses on processing recordings after they have been completed or imported. Although the final transcript is useful, it does not help the user understand the conversation while the meeting is taking place.
Real-time transcription would also support continuous dictation, accessibility workflows, noisy environments, and early detection of transcription errors.
Describe the solution you'd like
Add a
Live Modeto the NoJoin interface for real-time transcription.The interface should display the transcript continuously while audio is being captured. Ideally, it could support two display modes:
A possible workflow would be:
Useful options could include:
The feature should use local inference when NoJoin is configured for local processing.
Describe alternatives you've considered
Windows 11 Live Captions can provide real-time captions, but its transcript is not integrated with NoJoin and cannot easily become part of the same meeting, notes, speaker, and search workflow.
External captioning, dictation, and translation applications were also considered, but they create a separate workflow and require users to manually import or combine their results afterward.
Processing short audio chunks as separate recordings is another possibility, but it produces fragmented transcripts and is not suitable for a continuous meeting.
Additional context
The main use case is helping people follow meetings conducted in a language that is not their native language.
Other relevant use cases include:
The first implementation could prioritize sentence-by-sentence transcription. Word-by-word partial results could be added as an optional lower-latency mode if the selected transcription backend supports streaming or incremental decoding.
When the session ends, NoJoin should perform a final transcription pass if necessary so that the stored transcript prioritizes accuracy over the provisional live output.