Every voice, by name.

Nanoru transcribes your audio and video and puts a speaker on every line. It recognises people by their voice, so a presenter named once is named in every recording you send.

The Friday Panel, four episodes
  1. 213 This week Sam Okafor 43%, Unknown #8 41%, Alex Reid 11%, Music and adverts 6%
  2. 212 Last week Alex Reid 46%, Sam Okafor 45%, Music and adverts 6%, Unknown #9 3%
  3. 211 Two weeks ago Unknown #8 36%, Alex Reid 33%, Sam Okafor 24%, Music and adverts 6%
  4. 198 In March Unknown #8 36%, Sam Okafor 36%, Alex Reid 22%, Music and adverts 6%
  • Sam Okafor 4 episodes, 88 min
  • Alex Reid 4 episodes, 66 min
  • Unknown #8 3 episodes, 66 min
  • Unknown #9 1 episode, 2 min
  • Music and adverts

Nanoru has heard Unknown #8 in 3 episodes. At 0:59 this week the presenter introduced her as Maria Lopez.

This show is invented. Each bar is one episode, coloured by who is speaking.

Speaker 1 is a different person in every file

Most transcription services number the people in a recording, and the numbering starts again with each file. Speaker 1 in one episode is Speaker 0 in the next.

So finding every episode a guest appeared in, or the moment a host read out a sponsor’s name, still means someone listening back.

From a speech API
  1. 0:11 Speaker 0 Hello and welcome to The Friday Panel. I’m Sam Okafor.
  2. 0:15 Speaker 1 And I’m Alex Reid.
  3. 0:59 Speaker 0 With us this week is the chef Maria Lopez.
From Nanoru
  1. 0:11 Sam Okafor Hello and welcome to The Friday Panel. I’m Sam Okafor.
  2. 0:15 Alex Reid And I’m Alex Reid.
  3. 0:59 Sam Okafor With us this week is the chef Maria Lopez.

How it works

  1. Send your audio or video

    Upload files, point Nanoru at an archive, add a podcast feed, or connect a live stream.

  2. Nanoru names the voices it knows

    It transcribes the speech, works out who is talking when, and matches each voice to your voiceprints.

  3. Name a new voice once

    You name them with one key press in your dashboard. Nanoru then names that person in every recording, including the ones it has already processed.

You get results in your dashboard, as JSON over the API with webhooks, and as SRT or VTT captions that carry the names. What Nanoru does

What each line tells you

Each line has a speaker. Nanoru also marks adverts, intros and jingles that repeat between episodes, and lines where two people talk at once.

An unknown speaker gets one label, such as Unknown #8, and keeps it in every recording until you name them. Once Nanoru has been calibrated on a sample of your recordings that a person has checked, each line also carries a confidence score.

  1. 0:00Nanoru recognises the theme music, because the same 11 seconds open every earlier episode.
  2. 0:11 Sam Okafor Hello and welcome to The Friday Panel.Nanoru named him by his voice, with a confidence of 0.99.
  3. 0:17This is a 42-second advert. You can leave adverts out of exports.
  4. 1:03 Unknown #8 Thanks for having me back.Nobody has named her yet. The presenter’s introduction suggests Maria Lopez.
  5. 1:05 Alex Reid Third time now. You’re nearly a regular.He spoke over the end of the line before.
  6. 1:08 Unknown #8 Then I’d like a mug with my name on it.The confidence is 0.71, so this line goes on your review list.
These are the opening lines of the newest episode above.

How often it credits the right person

We ran Nanoru and five other systems on the same 9.8 hours of public recordings. The charts count, per 1,000 words a system transcribed correctly, how many it put under the wrong person.

ElevenLabs Scribe v3 did better than Nanoru on podcast interviews. Nanoru did better than all five on court hearings, where a dozen people interrupt each other, and better than the other four on podcasts.

The full results and how we measured them

Podcast interviews: words credited to the wrong person, per 1,000
  • ElevenLabs Scribe v3 1.2
  • Nanoru 2.1
  • Whisper large-v3 with DiariZen 2.8
  • Whisper large-v3 with pyannote 3.1 6.0
  • ElevenLabs Scribe v2 8.3
  • NVIDIA Nemotron-3.5-Transcribe 34.1
Court hearings: words credited to the wrong person, per 1,000
  • Nanoru 1.9
  • ElevenLabs Scribe v3 8.7
  • Whisper large-v3 with DiariZen 11.0
  • Whisper large-v3 with pyannote 3.1 23.1
  • ElevenLabs Scribe v2 26.9
  • NVIDIA Nemotron-3.5-Transcribe 66.4

The bars show words credited to the wrong person, per 1,000. Shorter is better. These are our own tests from November 2026, on recordings Nanoru’s settings were chosen on.

32p to 75p for an hour of audio

Recordings are covered by monthly plans, and live streams are charged by the hour. The table compares Nanoru with Audioscrape, which also keeps voiceprints, and with ElevenLabs Scribe, the speech API we tested it against.

Currency
List price for an hour of recorded audio, and how each product names speakers
ProductHow it names speakers Price on a scale from 0 to £1.50An hour of audio
NanoruBy voice; voiceprints kept for youBy voice; voiceprints kept for you32p to 75p
AudioscrapeBy voice; voiceprints kept for youBy voice; voiceprints kept for you98p to £1.37
ElevenLabs Scribe v2 and v3Numbered, not namedNumbered, not named15p
These are published list prices from November 2026, converted to pounds. Nanoru’s range covers its monthly plans. Human transcription with speaker names costs £36 to £115 an hour.

Plans, live streams and the cost for your hours

Test it on your own recordings

Send 20 episodes or 20 hours of recordings. We return them with your presenters named, at no charge.

Ask for a free trial