HeartMuLa Transcribe Review: Lyrics Transcription without Timestamps
Lyric text without timestamps, confidence values or speaker labels.
HeartMuLa Transcribe extracts sung lyrics from an audio file and returns them as text. One audio field, no options, multilingual, billed per transcription rather than per minute.
- Lydia
- Tested Sep 30, 2026
- Updated Oct 7, 2026
The verdict
We did not create new paid samples for this article. Published examples remain examples of a particular input and output, rather than a statistically controlled comparison. The illustration accompanying this article is editorial artwork, not a sample produced by the reviewed model.
What we liked and what held it back
What we liked
- Music catalog metadata. Generate lyric text across a catalog to improve search, display and tagging. Per-run pricing makes the cost of a backlog a simple multiplication instead of a function of total runtime.
- Subtitles for music-backed video. Pull the sung lines out of a track so they can be timed into captions for short-form video, where the vocal sits over a dense mix.
What held it back
- One audio URL in, lyric text out, nothing to configure: no language hint, no timestamps, no speaker labels, no confidence values and no choice of output shape. The model is tuned for sung vocals over music; heavily layered mixes, whispered or screamed vocals and very low-bitrate sources degrade the result, and instrumental audio returns little or nothing. The URL must resolve directly to an audio file — a page that returns HTML fails as an invalid asset. This endpoint produces no media asset, so nothing from it appears in the 14-day media retention.
What the evidence shows
Our existing 30 September 2026 diagnostic used six LibriSpeech test-clean clips, six speakers, 53.6 seconds of English read speech and 146 reference words. HeartMuLa Transcribe recorded 4.11% word error rate: 4 substitutions, 0 deletions and 2 insertions. The transcription methodology explains normalization and the limited task being measured. These are documented earlier results, reused here without a new transcription request.
This small, clean, short-form sample does not establish accuracy on meetings, overlapping speakers, noise, other languages or sung lyrics. Word error rate measures word differences against a reference; it does not measure punctuation, formatting or speaker attribution. Keep that distinction when choosing a tool for subtitles, a searchable archive or a final transcript.
A practical starting brief
Use an audio file that you own or have permission to process. Check the beginning and ending, one proper name and a difficult passage against the audio. For a noisy recording, keep a written record of words you cannot verify instead of silently treating the transcript as authoritative. The model page explains the supported file input.
Start with the material that determines acceptance: a name, a difficult number, a sentence with an emotional turn, or the passage you cannot afford to misread. Listen or read against the original, including the ending. Keep the input fixed while changing one control, so a difference in wording does not get mistaken for a better model. A pleasant voice alone does not prove that every word survived.
Read the current quote on the model task page before submitting. Per-character, per-second and per-request billing cannot be compared by the displayed starting number alone. Compare the same passage or recording, and include any format conversion or alignment you still need to do.
How to choose an alternative
Choose by the missing capability before comparing a headline quality number. An alternative should remove a concrete bottleneck—voice selection, a longer script, delivery direction or the right kind of transcription—rather than simply have a newer name. The current pages below show their actual tasks and price units.
What HeartMuLa Transcribe can do on SpicyAPI
Straight from the live model specification on SpicyAPI. Some creators offer more in their own apps; this is what you can call through our API today.
| Feature | Speech to text |
|---|---|
| Seed control | No |
| Negative prompt | No |
What do you think?
Was this review useful?
Log in to vote.
Comments
Comments are checked automatically and by our team. We remove spam, offensive content, personal attacks and anything that names a real person without consent. Report a comment and we review it within one business day.
Tried this model? Share what worked. You need a free account to comment, so we can keep spam out.
HeartMuLa Transcribe FAQ
Try HeartMuLa Transcribe yourself
Set every parameter in the Playground and see the exact price before you run it.
More reviews
See the leaderboards →
Grok STT
One audio file per request, 1 second to 2 hours.
Tested Sep 30, 2026
Read the review
Gemini 3.8 Flash-Lite TTS
One or two voices per request: up to 5,000 characters of script for one voice, or up to 20 dialogue lines of 250 characters each for two.
Not hands-on tested · catalogue checked Oct 7, 2026
Read the review
Gemini 3.8 Flash TTS
One or two voices per request: up to 5,000 characters of script for one voice, or up to 20 dialogue lines of 250 characters each for two.
Not hands-on tested · catalogue checked Oct 7, 2026
Read the review