Transcripts and English translations you can check against the audio.

Transcribe

Drafts that show where they’re unsure.

Each language uses a speech model chosen for it, including a model fine-tuned for Bangla. English mixed into any language is expected.

• ✓Finds who is speaking; you name each speaker.
• ✓Unsure words underlined as a hint to listen again.
• ✓When the model gets stuck repeating itself, that stretch is decoded again from the audio. For Bangla, words cut where the model’s 30-second windows meet are recovered the same way.
• ✓Progress you can trust: a real progress bar with time remaining.
• ✓Bring transcripts you already have: .txt, .docx, .srt or .vtt (Research).

Translate

Translations that keep the speaker’s meaning.

Every non-English interview is translated into English by a language model running on your computer, passage by passage, so tone, sarcasm and register come through.

• ✓Bangla idioms translated by meaning, not word for word.
• ✓Right-click a phrase for other translations.
• ✓Retranslate a single passage after you correct it.
• ✓Translation memory: fix a term once, and it is used in every translation after (Research).

• ✓English above the original, with speaker and time.

• ✓A list of places the recogniser was unsure, each one click away.

• ✓Search the interview, typing in Bangla or English.

• ✓Mark passages reviewed, so you know what you’ve checked.

• ✓Summaries in the interview’s language or English, with key moments linked to the audio.

Languages

Eight languages, translated into English.

Accuracy depends on the language, the recording and how casually people speak; casual Bangla is hard for any speech model, so expect more to correct. Two non-English languages in one recording are not supported.