Transcribe audio to text
Upload a recording, an interview or a video, or paste a link, and transcribe it to text with who said what and when, ready to copy or download as subtitles.
- Up to 99 languages
- Speaker labels
- TXT, SRT and VTT
- 30 credits an hour
Speaker labels and timestamps are included. Uploads are deleted after 7 days.
Made with MakeMidas
Real results from the models on this site, with the prompts that made them.
0:00Speaker 1:欢迎收听城市早班车,我是主持人林晓,今天我们去杭州,
0:06Speaker 1:聊一家开了30年的面馆。
0:09Speaker 2:大家好,我是记者周明。
0:11Speaker 2:这家店叫老陈面馆,每天早上6点开门,一碗牛肉面卖28块。
0:17Speaker 1:听说排队的人特别多。
0:20Speaker 2:对,周末要排40分钟。
0:22Speaker 2:老板陈阿姨说,秘诀就是那天凌晨4点开始熬的骨头汤。
0:27Speaker 1:感谢周明的报道,我们明天见。
0:00Speaker 1:Welcome back to Market Minute. I'm Sam, and today we're talking about the tiny bakery that sold out of croissants before sunrise.
0:08Speaker 2:Thanks, Sam. The owner told me the secret is simple: bake half the batch at four in the morning, and the rest at six, so the second wave is still warm when the commuters arrive.
0:20Speaker 1:So timing beats volume?
0:22Speaker 2:Exactly. They sell fewer pastries than last year, but nothing goes to waste, and the line starts before the doors even open.
0:30Speaker 1:That's Market Minute. See you tomorrow.
Popular models
0:00Speaker 1:欢迎收听城市早班车,我是主持人林晓,今天我们去杭州,
0:06Speaker 1:聊一家开了30年的面馆。
0:09Speaker 2:大家好,我是记者周明。
The most accurate Whisper, strongest on non-English audio and names. About 99 languages, detected automatically, with speaker labels and timestamped segments.
30 credits per hour of audio
0:00Speaker 1:Welcome back to Market Minute. I'm Sam, and today we're talking about the tiny bakery that sold out of croissants before sunrise.
0:08Speaker 2:Thanks, Sam. The owner told me the secret is simple: bake half the batch at four in the morning, and the rest at six, so the second wave is still warm when the commuters arrive.
0:20Speaker 1:So timing beats volume?
NVIDIA's speech model: fast, short files finish in seconds. About 40 languages, with speaker labels and timestamps.
30 credits per hour of audio
Meetings, interviews and podcasts
Upload the recording and read it back as text. Each turn is labelled with its speaker, so a two-person interview or a panel stays easy to follow.
Subtitles for your videos
Download SRT or VTT files with timed captions that drop into YouTube, Premiere, DaVinci Resolve or a web player. Lines are split at pauses and sentence ends.
Audio or video, file or link
MP3, WAV, M4A, AAC, FLAC, OGG, Opus, WEBM, MP4 and MOV files are accepted. Drop a file from your computer or paste a direct link to one.
See the price before you start
The length is read as soon as the file is uploaded, and the button shows what it costs. If the transcript comes back shorter, the difference is refunded; if a file's header understates its length, the real length is charged.
Keep creating
One account, one credit balance for images, video and audio.
Choose your plan
A monthly plan or a one-time credit pack gives every model and clean output.
Black Friday: 1,250 credits
- Base credits: 1,000
- Black Friday bonus: 250
- Valid for 12 months
Black Friday: 6,250 credits
- Base credits: 5,250
- Black Friday bonus: 1,000
- Valid for 12 months
Black Friday: 12,500 credits
- Base credits: 11,000
- Black Friday bonus: 1,500
- Valid for 12 months
Black Friday: 31,250 credits
- Base credits: 28,750
- Black Friday bonus: 2,500
- Valid for 12 months
Subscribers get 62,500
- Base credits: 62,500
- Valid for 12 months
Any pack unlocks every model and several outputs at once, even without a subscription.
Questions
How do I transcribe audio to text?
Upload an audio or video file above, or paste a direct link, and choose a model. When the transcript is ready, copy the text or download it as TXT, SRT or VTT, with speaker labels and timestamps.
Which model should I choose?
Whisper Large v2 is the default and the most accurate Whisper model we run: it is strongest on non-English audio and names, and covers about 99 languages with automatic language detection, speaker labels and timestamped segments. Nemotron 3.5 Transcribe is NVIDIA's speech model: it is fast, finishing short files in seconds, and covers about 40 languages, also with speaker labels and timestamps. Both cost the same per hour.
Which languages can it transcribe?
It depends on the model: Whisper Large v2 covers about 99 languages and Nemotron 3.5 Transcribe about 40, both including English, Spanish, French, German, Portuguese, Italian, Japanese, Korean and Chinese. The language is detected from the audio, so there is nothing to pick.
How big and how long can a file be?
Up to 100 MB and 60 minutes per file. For a longer recording, split it into parts first, or export the audio track of a video as MP3 or M4A to make the file smaller.
Which file formats work?
MP3, WAV, M4A, AAC, FLAC, OGG, Opus and WEBM audio, and MP4 or MOV video with a sound track. Whisper Large v2 reads MP3, M4A, WAV, MP4, WEBM and OGG; choose Nemotron 3.5 Transcribe for AAC, FLAC, Opus or MOV. A link must download the file directly, without a login.
How much does a transcript cost?
30 credits per hour of audio, charged by length: 1 credit covers each started two minutes, and every file costs at least 1 credit. Credits come from a plan or a credit pack, and a failed transcript is refunded.
What happens to my files?
Uploads are used only to make your transcript and are deleted automatically after 7 days. The transcript stays in your history until you no longer need it, and only you can open it.