Upload any audio, get an accurate transcript with timestamps.
15 free minutesNo sign-upSRT export
MP3, WAV, M4A, MP4, FLAC, OGG · Up to 500 MB
What is audio transcription?
Audio transcription converts spoken words in a recording into written text. Get Audio Transcript accepts MP3, WAV, M4A, MP4, FLAC and OGG files and produces transcripts with timestamps. Review the result against your recording, then export it as text or subtitles.
LISTEN & READ
Hear it. Read it back.
An interactive audio sample
AUDIO
Shoreline field notes
Narrated sample · WAV
00:00 / 00:09
TRANSCRIPT
SOURCE FILES
From voice memo to video.
MP3WAVM4AMP4FLACOGG
Convert WAV to text from an original recorder file to preserve speech detail. For M4A to text, upload the original phone recording instead of a compressed chat attachment. To turn a voice memo to text, share it from your recorder and save the audio file first. MP4 video, lossless FLAC and OGG recordings work here too. Each paid upload can be up to two hours or 500 MB, whichever comes first.
From recording to transcript.
01
Upload the recording
Choose a file or paste a direct media link.
02
Transcribe the speech
Select the spoken language or let the service detect it.
03
Export your work
Download TXT, DOCX, SRT, VTT or JSON.
IN PRACTICE
What are you recording?
Meetings
Revisit the decision
A useful meeting transcript lets you locate the moment a proposal became a decision. Start with the sentence where someone agreed to do the work. Check the minute before it for a condition, a deadline or a dissenting voice. Then write the owner, action and date in your notes. Keep the timestamp beside that line so a colleague can hear the exchange without replaying the entire call. For a remote meeting, choose the original recording from the host whenever possible and check whether everyone on the call knew the recording was being made.
An interview can sound very different on the page. Listen to the question immediately before a strong answer and keep the timestamp with the quote as it moves into your draft. For a podcast, mark the point where the topic changes and use those markers as the beginnings of your show notes. If two guests talk over each other, revisit the original audio before assigning a sentence to either person. Save a verbatim copy first, then edit a separate reading copy so the producer can always trace a cut back to the recording.
Turn a recorded lecture into notes you can search before an exam. Locate the first definition of a new term, then listen to the example that follows. Write the definition in your own words and leave a time reference next to it. Slide titles, formulas and names often need a second pass because the speaker may point to something the microphone never captured. A short guest upload works well for testing one difficult explanation before processing a full class. Get permission before sharing a recording or transcript with classmates.
A written transcript gives you a starting point for captions, a newsletter excerpt or a rough cut. First identify the passage that can stand on its own. Play it with the previous sentence to check whether a pronoun or a reference loses its meaning. Export subtitle cues and open them beside the final edit. If you remove an introduction or change the playback speed, align the captions to the edited timeline before publishing. Save the plain text as well so your next description, article or social clip uses the same spelling for names and places.
When audio transcription saves you a second listen
Someone remembers a phrase from an hour-long recording but cannot remember who said it or when. A transcript gives you a way to search for a distinctive word, find the surrounding passage and return to the sound. Keep the timestamp with the excerpt as it moves into an article, a lesson plan or meeting notes. If someone challenges the wording, you can play the relevant exchange without asking the whole group to listen again.
The useful export depends on who opens it next. A writer can search plain text across a folder of interviews. A reviewer can open DOCX to make corrections and leave comments. A video editor needs cue times to see when a line appears and disappears. Downloading both a reading copy and a timed copy at the end of a session lets those people work from the same version of the speech.
Which language should I choose?
The recording menu lets you set English, Mandarin Chinese, Spanish, Portuguese, German, French, Italian or Indonesian. Select the language being spoken, even if the person reading the transcript speaks another language. Use automatic detection if you are unsure. For a bilingual interview, listen to a section where the language changes and review the surrounding sentences carefully. A good result from the opening minute tells you little about a switch later in the conversation.
Keep the spelling of people, streets and organizations in the language the speaker used. If an audience needs a translation, retain the original transcript as a reference while preparing it. That gives a bilingual reviewer a place to check a technical term, a joke or a phrase with more than one possible meaning.
Three things to check before sharing
First, read a name against a source that spells it out. A sentence can sound right while referring to the wrong person. Next, replay numbers with their units attached. “Fifteen” in one sentence and “per week” in the next belong together if that is how the speaker described a rate. Finally, check the timeline against the file your reader will actually play. An editor who cuts the introduction from a recording also moves the point where every later quote begins.
Give the next reader a usable reference
A timestamp works best with a filename. Suppose two teammates have copies of a talk, one beginning with the moderator’s welcome and one beginning with the speaker’s first slide. A note that says “the example starts at 04:10” sends the readers to different sentences. When you share a passage, include the recording title, the version you transcribed and the time where the thought begins. A short text excerpt helps the reader confirm they have reached the intended place.
For a long recording, use a few descriptive phrases as an index in your own notes. A label such as “guest explains why the launch moved” will be easier to find than “discussion continues.” Keep the labels outside the quoted text, with a timestamp beside each. You can then hand a colleague a short reading list while preserving a full transcript for someone who needs to check the context later.
Accuracy and privacy
Our transcription engine is AssemblyAI Universal-2. See AssemblyAI’s supported languages for model-specific coverage. Results depend on the speaker, microphone and background audio. In a launch check using four English audiobook clips from four speakers, word accuracy ranged from 90% to 100% per clip. Across 95 reference words there was one substitution, “Oh” for “O.” We ignored case and punctuation and scored insertions, deletions and substitutions. These were clean readings; calls with overlapping voices need their own review. LibriSpeech test-clean corpus by Panayotov and colleagues, licensed under CC BY 4.0.
Audio is sent over HTTPS and mixed to a single channel for transcription. Local working files are removed after submission. Completed transcripts are scheduled for removal once 24 hours have elapsed since submission. Use Delete on a completed result to request earlier removal at the provider. Keep your own exported copy. Read our privacy policy and AssemblyAI’s data retention documentation.
Each guest can transcribe up to 15 minutes in total while the monthly free pool has capacity.
Which languages can I use?
Use automatic detection or select the language you recorded. The language menu includes English, Mandarin Chinese, Spanish, French, German, Portuguese, Italian and Indonesian.
Can I download SRT subtitles?
Yes. Export SRT for numbered subtitle cues or WebVTT for a web player. Each cue has a start and an end time.
How well does it handle Chinese?
For Mandarin recordings, select Chinese if you know the language. Listen back to names, numbers and passages that switch between Mandarin and English.
Can I send several recordings or use an API?
Plus and Studio include batch uploads. Studio also includes API access. Batch files and API requests use your plan’s shared minute balance.
How large can a recording be?
Paid recordings can run up to two hours and 500 MB each. A guest recording must fit within the remaining trial minutes.
What happens to my recording?
Temporary local audio is removed after submission. Download your result when ready. Completed transcripts are scheduled for deletion after 24 hours. You can also delete a finished transcript from the result view.
Do I need an account?
You can try a recording as a guest. Paid minutes belong to the email address used at checkout.
Can I use the output commercially?
You can export a transcript for a commercial project when you have permission to use the underlying recording.