How to turn audio into text

Which AI transcription tools turn Latvian speech into text?

Each cell below comes from the tool's own page, as it read in October 2026. Tools change quickly, so check the page before you rely on a detail.

For Russian and English the choice is much wider: most of the tools above handle them, as do many others, while the list for Latvian stays short.

How accurate is Latvian speech recognition, and what lowers it?

Accuracy depends on the recording at least as much as on the tool, and Whisper's own page says its performance varies widely from language to language. These lower it:

• background noise and echo;
• too much distance from the microphone;
• people talking over each other;
• fast or mumbled speech;
• dialect;
• names, brands and jargon;
• switching between Latvian, Russian and English, which is common in conversations in Riga;
• music under the speech.

Before you send a long file, test five minutes of your own recording in two tools and compare the results.

How do you record so the transcript comes out clean?

1. Give each speaker their own microphone, close to the mouth.
2. Record in a quiet room with little echo. The guide to improving audio quality shows how to find and fix the usual problems.
3. Agree before you start that people will not talk over each other.
4. Keep music out of the recording and add it in the edit.
5. Record a lossless WAV file, or a high-bitrate one, and give the tool the clean speech track rather than the finished mix.
6. Say the names and terms clearly at the start. It helps your own proofreading later.

At our studio in Riga, an audio podcast recording gives each person at the table their own professional microphone, and the sound is mastered.

How do you clean up an interview transcript and label the speakers?

After transcribing an interview, always proofread the text:

1. Make a first pass for names, numbers and terms. Use find and replace for an error that repeats.
2. Label the speakers. Some tools separate them automatically: according to their pages, Tilde and Word's Transcribe both do, and Word numbers them ("Speaker 1", "Speaker 2"). Otherwise mark them by hand, for example "Host" and "Guest".
3. Add timestamps at each change of speaker or at each paragraph, so you can find a passage in the audio.
4. Choose the style: full verbatim or clean verbatim (see the table below).
5. Listen through once at 1.25 or 1.5 times speed while you read along.

What can you make from a transcript?

• show notes and a summary of the episode for when you publish your podcast;
• chapters with timestamps;
• an article or a blog post;
• quotes for social posts;
• captions as an SRT file;
• the best moments for short clips, found by reading instead of listening;
• a searchable archive of past episodes.

For subtitles or dubbing in another language, see video translation and dubbing.

If you would rather hand over the edit itself, our editors edit podcast video and sound for €40/h, including recordings made outside the studio.

Is transcription free, and what happens to your recordings?

Options without a subscription:

• Google Docs voice typing, for live speech;
• YouTube's automatic captions, on a video you upload: once you have corrected them, you can download the subtitles in YouTube Studio under Subtitles, then Options and Download subtitles;
• Whisper on your own computer: it is open source under the MIT licence, which allows use free of charge, but the larger models need a capable machine.

Word's Transcribe comes with a Microsoft 365 subscription and covers 300 minutes of uploaded audio a month. Tilde has monthly plans with a 14-day free trial, as listed in the table above.

Privacy depends on where the audio is processed. Whisper on your own computer keeps the audio with you. Cloud tools process it on their servers, so read what each page says about storage and use. Tilde's page, for example, says your data is not used to train public AI models and offers on-premises, private-cloud and EU-based options. Word saves uploaded recordings to a Transcribed Files folder in OneDrive. Ask interviewees for permission before you upload their voices, and keep confidential recordings away from tools whose terms you have not read. This is practical advice, not legal advice.