Descript: Amazing improvements in transcription + Transcription Glossary in May 2026

I have been a very happy user of Descript since 2023 at least, although I have mainly used it for audio, less frequently with video. I have lost count of how many articles I have published about Descript over the years, as it kept on adding valuable features. Most of my Descript articles have been very positive. The only negative I can recall was in 2024: Descript Transcription Glossary: 3 failures. Even though it was beginning to seem like they might never fix the issues, I am ecstatic to inform you that those three issues have now been solved, in addition to the general quality of the transcriptions has improved, both in Castilian (castellano) (aka «Spanish/español») and in English transcriptions. Ahead, I will share screenshots to illustrate the quantum leap transcription improvement from a partial portion of Block A of my CapicúaFM show, between the April and May episodes. I’ll also clarify how the underlying improvement actually came from Descript’s partner for transcription: ElevenLabs (which I have also covered in several articles). Finally, I’ll demonstrate about how the Transcription Glossary still shows an absurd error warning, which fortunately no longer matters.

To recap the three failures covered in my 2024 article

The three failures covered in that article were:

1. Brands with built-in numbers were prohibited/ignored in the Transcription Glossary in Descript. For example, popular names like 21st Century Fox, 23andMe, 2Wire, 3Com, 3M, 7-Eleven, 7up, A1 Steak Sauce, Boeing 747, Century 21, Cloud9, Forever 21, Levi’s 501 and Motel 6 were prohibited and ignored, as well as being poorly transcribed.
2. Accent marks were prohibited/ignored in Descript’s Transcription Glossary, even if the content was in English. Those included accented names like: Bacardí, CapicúaFM, Déjà vu breakfast company, El País, elPeriódico, Medéa, Nescafé, Nestlé, Piña Colada, Résumé writing service and The Piñata Factory.
3. Descript’s Transcription Glossary was «off limits» outside of English content.

Quantum leap improvement in May 2026-use of Transcription Glossary

Before May 2026, the Transcription Glossary was like a non-functioning prop in a movie set or theatrical play. Prior to May, this was the typical raw transcription of Block A from my CapicúaFM:

Above, you will see all of the errors in the partial raw transcription from episode 109 from April 2026:

1. Víctor Martorella‘s first and last names were misspelled, even though his properly-written names were in my Transcription Glossary, just ignored.
2. Victoria Mesas García’s first name and first surname were inappropriately transcribed in lowercase (since they also happen to be dictionary words), even though her properly-written names was in my Transcription Glossary, just ignored.
3. Andreína Ascanio Toro‘s first name and both of her surnames were transcribed inappropriately, even though her names were in my Transcription Glossary, just ignored.
4. My compound word queridoyente was transcribed as the two separate words querido and oyente, even though the compound word queridoyente was in my Transcription Glossary, just ignored.
5. CapicúaFM was completely botched, even though it appeared correctly in my Transcription Glossary, just ignored.
6. TecnoTur was completely botched, even though it appeared correctly in my Transcription Glossary, just ignored.
7. My name (Allan Tépper) was reduced to just «al», even though it appeared correctly in my Transcription Glossary, just ignored.

Now, let’s see the improvement in May of 2026.

Above, you’ll see the partial raw transcription from episode 110 from May 2026.

1. Víctor Martorella‘s first and last names were transcribed perfectly, including the accent mark on his first name, since the Transcription Glossary is finally used!
2. Victoria Mesas García‘s was transcribed perfectly, since the Transcription Glossary is finally used!
3. Andreína Ascanio Toro‘s first name and both of her surnames were transcribed perfectly, since the Transcription Glossary is finally used!
4. My compound word queridoyente was transcribed perfectly, since the Transcription Glossary is finally used!
5. CapicúaFM was transcribed perfectly, since the Transcription Glossary is finally used!
6. TecnoTur was transcribed perfectly, since the Transcription Glossary is finally used!
7. My name (Allan Tépper) was transcribed perfectly, since the Transcription Glossary is finally used!
8. María Esnoz‘s name is transcribed perfectly, since the Transcription Glossary is finally used!

Quantum leap improvement in May 2026-use of numbers

I just made a brand- new audio recording of the same brand names with numbers I had used for the 2024 article. Without even listing any of the following brand names in my Transcription Glossary now, the results are great.

Considering that now, I don’t have any of those brand names listed in my Transcription Glossary, I think Descript did a wonderful job.

Improvements partially thanks to ElevenLabs

Apparently, part of the amazing improvments I just documented are thanks to Descript‘s partner, ElevenLabs. I say that since I found this in the changelog for May 2026:

Improved transcription with ElevenLabs Scribe v2
Scribe v2 is now the default transcription model for all supported languages. It outperformed our other models in testing, resulting in fewer corrections per minute and fewer capitalization corrections.
What’s changed:

• Automatic mode now uses Scribe v2 (previously Rev v2 for English and now Scribe v2 for all languages)
• The model selector now displays the actual model names: Automatic, Scribe v2, and Rev v2
• UK English: Scribe v2 uses US English spelling. To get UK spelling, select Rev v2 in the model selector

I suppose we can thank both Descript and ElevenLabs!

The absurd error warning which persists in the Transcription Glossary window in Descript

As you’ll see in the above screenshot (which only shows a portion of the terms I have listed in my Transcription Glossary). It is absurd since none of my current terms include any numbers or numerals. I suspect that it may be a leftover false positive from back when Descript officially banned accent marks in the Transcription Glossary, even though accent marks are not numbers… and the accented terms now finally work perfectly, I really don’t mind if Descript takes another two years to correct this absurd error warning, as long as the transcriptions continue to be so pristine as they are after this wonderful May 2026 upgrade.

Lee este artículo en buen castellano

Descript: Impresionantes mejoras en la transcripción + Glosario de transcripción en mayo del 2026