Post production toolkit · multitrack rushes
Every line, on the person who said it.
Upload the day’s multitrack audio as it was recorded. Get back a transcript for every microphone, the conversations between them, and a cut you can hand to the edit.
Free trial
Two channel-hours of processing - a ten-minute scene from a ten-mic show, or an hour of a two-mic interview - one seat, twenty-five assistant questions, for thirty days. A card on file, and nothing charged: it is how one trial stays one trial. Buying a package later is a separate, deliberate step.
Your own material, not a demo. When you are ready, a package is £1.80£2.16 a channel-hour ex VATinc VAT, prepaid, and that is the whole price list.
- Self-hosted models
- No AI vendor sees your material
- Data stored in the EU
- Any cast size
Every microphone, transcribed on its own
Every radio mic on the shoot gets its own transcript. Spill from the rest of the room is thrown away, so each line lands on the person who actually said it. Then the mics are read together, to work out who was in conversation with whom, and when.
Hear one conversation, not the whole room
Follow one person through the day. The transcript shows only the conversation they are in, and playback is only the microphones in it. When they leave one group and join another, the page and the sound go with them.
withMarek Nowak, Harriet Holloway, Des Clarke, Owen Castell, Lottie Marsh, Georgia Tate, Tunde Adeyemi
withOwen Castell, Lottie Marsh, Des Clarke
withDes Clarke
Find every mention, and play it
Type a word. Every time it was said comes back with who said it and when, and a play button on each hit. Narrow it to one speaker or one day, then open the transcript at that line.
Right, okay. Can I say something first? I voted for Yasmin, and I was wrong.
I wonder. We were all so sure about Yasmin last night. I would hate us to be sure of the wrong person twice.
At work I watch people get bad news. A real shock comes slowly. Yasmin accused him, and nothing arrived.
At the Gate she said, check his sums.
Since Yasmin stood at the Gate and said, check his sums. So I checked.
Here is the notebook. Day three, the linen room. Linen: two hundred and twelve.
I was on the stairs with the wine total. You called up: linen, two hundred and twelve. I heard it.
Ask what was said, and who was listening
Ask in plain English, across the whole production: who said what, to whom, and when. Every answer quotes the lines it rests on, with the timecode and a play button. If the rushes do not support an answer, it says so rather than making one up.
Marek, shut the door a moment, would you?
The count was wrong on day three.
Pass the pull to the edit
Pull a line wherever you find it: a transcript row, a search hit, an assistant answer. Put the pulls in order and hand the cut over as an EDL, an ALE, an AAF, an FCPXML or a marker list. Or copy one line with its timecode and paste it into a shared doc or a WhatsApp thread.
“Right, okay. I voted for Yasmin, and I was wrong.”
“At the Gate she said, check his sums.”
“The count was wrong on day three.”
“At the Gate she said, check his sums.”
“Right, okay. Can I say something first? I voted for Yasmin, and I was wrong.”
“Tunde. I watched his face when Yasmin accused him. He did not look hurt. He looked ready.”
“The count was wrong on day three.”
“Here is the notebook. Day three, the linen room. Linen: two hundred and twelve.”
“I was on the stairs with the wine total. You called up: linen, two hundred and twelve. I heard it.”
“Marek, that was two days ago. Why did you never say?”
Confidentiality
Your rushes stay yours.
Nobody else’s AI.
Transcription, speaker identification and the assistant all run on models we host. No AI company receives your material, and nobody trains on it.
Access is enforced in the database.
Per-production roles, invitations and two-factor sign-in, applied one layer below the application, so every export and every media file inherits them.
Source audio can be purged on request.
Transcripts, search and exports stay, and every export relinks to your own media in the edit.
Consent
Who is speaking: the production decides.
Two ways to put a name on a line. Either way you get spill-clean transcripts with the conversation structure on top.
Voice printing
Enrols a few seconds of each contributor’s speech, under a signed release, and checks every line against it.
Channel ownership
Takes the names from the track names your recordist already logs, and stores nothing biometric at all.
On the floor
Nothing changes.
Upload the multitrack audio exactly as it was recorded, with the mic names already on it. No new kit, no plug-in in the edit suite, no change to how you record.
Price
One price.
Charged per channel-hour: one recorded hour on one microphone. Everything is in it: everyone who needs an account, every export, the assistant, and your transcripts and working audio kept for two years. Quoted per production from your cast size and shoot length, with a not-to-exceed.
The one add-on is a dedicated GPU for the assistant, by the month: your own model server, kept warm through your working hours or around the clock, with no limit on questions.
Judge it on your own rushes day.
The fastest way past a reasonable scepticism about AI transcripts is to run it on material you already know. Send one day of multitrack audio as recorded, and compare the result with the transcript you would otherwise have paid for.
The free trial is that day, in miniature: two channel-hours of your own multitrack, one seat, twenty-five questions, thirty days. A card on file, nothing charged.
A fuller account of what the platform does, how it works, where it fits in your workflow, and how it is priced. Technical and security documentation - the processing architecture, the subprocessor list, and the answers your broadcaster’s security team will ask for - is available on request.