Transcription That Streams As It Listens

April 21, 2026

Drop in a 40-minute podcast and you used to stare at a frozen "Transcribing…" label for minutes before a wall of text appeared. Now the transcript fills in segment by segment as the audio is processed, with a real progress bar and a stage label that tells you which phase the system is in.

What's better

• The transcript fills in live. Every segment streams into the chat as soon as it's decoded — no more long silent waits on long files.
• The processing message tells you what's happening — analysing speech, decoding, diarising — so you know the system is alive, not stuck.
• A real progress bar, driven by seconds of audio processed, replaces the old spinner.
• You can scroll the partial transcript while it's still running and start reading before the final file lands.

When you'd notice it

• A 35-minute investor call recording. You used to wait three minutes for the whole thing to drop. Now you see the first 20 seconds almost immediately and can start scanning for key quotes while it finishes.
• A noisy field recording where you're not sure the system is coping. The stage label flips through "analysing speech → decoding → diarising" so you can see where the work actually is.
• Batch-transcribing several files. Because segments stream in live, you can tell within 10–15 seconds whether the language was detected correctly and cancel a wrong one early.

Try it

• "Transcribe this audio and summarise the key points."
• "Transcribe this meeting recording with speaker labels and timestamps."
• "Pull a transcript of this podcast and extract every quote longer than 20 words."

Heads up

• The final transcript is unchanged. You get the same complete text at the end — the only difference is you watch it build instead of waiting.
• Speaker labels may correct themselves near the end of a long file. That's the diarising pass catching up — expected, not a bug.
• CPU transcription got faster too. Short files finish quicker and long files feel alive throughout.