How to transcribe audio to text with MyClawn Transcribe

Step 1: upload your files

Use the upload box on the home page. Accepted: MP3, WAV, M4A, AAC, OGG, FLAC, MP4, MOV, WebM and MKV. Each file can be up to 4 GB and one job can hold up to 10 files. If the audio is inside a video, upload the video as is; there is nothing to convert or extract.

Step 2: say who is talking and how you want it

Write a short brief in plain words. Useful things to include:

• Speaker names, for example: Dana (host) and Luis (guest). Without names you get Speaker 1, Speaker 2. Names are never invented.
• Verbatim (keeps every um and false start) or clean read (default, tidies those out)
• Timestamps: every speaker turn by default, or every minute if you prefer
• Extras: a summary, action items, chapters, SRT or VTT subtitles
• Spellings of names, products and jargon that the recording may get wrong

Example brief: Two speakers, Dana and Luis. Remove ums. Timestamps every minute. Add a summary at the top.

Step 3: pay and wait

Checkout is a Stripe subscription: $25 a month for 20 files or $79 for 100. The job then runs in two passes. Speech recognition produces timed text, then an editing pass adds punctuation, paragraphs and speaker labels, and checks that the final timestamp is near the end of the recording.

Step 4: get your document

You get an email and a download page with:

When you upload several files, each output is named after its input file. A short job summary lists the recording duration, the speakers found and how many passages are marked [inaudible].

Prepare the recording for a better transcript

The transcript can only be as good as the audio. A few minutes before you upload make a real difference:

• Upload the original file, not a re-compressed copy. Larger files are fine, up to 4 GB.
• If each person was recorded on a separate track, upload each track as its own file and tell us which is which. Separate tracks avoid the crosstalk problem.
• Put the names and spellings of unusual words in the brief: company names, product names, people, places, technical terms.
• Say how many speakers there are and who they are. If a speaker only says a few words, say so.
• Tell us what the transcript will be used for (a published interview, internal notes, a research appendix). That decides verbatim or clean, and how much to tidy.

A real test run

We tested the service on a 4-minute recording. It returned a Word transcript in 80 seconds. That is one small run, not a promise about every file: longer recordings take longer, and we do not quote a guaranteed turnaround. Typically the result arrives well under an hour.

Briefs that work, by situation

Changes

If a label is wrong or a name is misspelled, reply in one line. Each job includes two rounds of changes. If it still misses, you have 14 days to ask for a refund.

Limits you should know before you upload

• It is AI, not a person listening to every word. Heavy accents, crosstalk and noisy rooms will have some misses. We flag what we cannot hear instead of guessing.
• Speaker labels are inferred from the conversation and the names you give. On calls with many similar voices, check the labels.
• No certified legal or court transcripts.
• No medical records under HIPAA. Please do not upload them.
• It does not translate. Each recording is transcribed in the language spoken.
• Files are deleted after 30 days.

What happens with long or several files

A job can hold up to 10 files. Each file produces its own transcript, named after the input file, so ten interviews come back as ten documents instead of one merged blob. Long recordings simply take longer to process. After processing, the last timestamp is compared with the length of the recording, so a transcript that stops halfway is caught before delivery. If you want one combined document, say so in the brief.

Questions

More about Transcribe