<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/extremely-fast-and-accurate-transcription-with-granite-speech-5-0-turbo-ctc-ivnplcd09" -->

---
title: Extremely Fast and Accurate Transcription with Granite...
description: IBM released two new 470M-parameter English speech recognition models in the Granite Speech family: granite-speech-5.0-470m-turboctc (Apache 2.0) and...
canonical: https://daily.dev/posts/extremely-fast-and-accurate-transcription-with-granite-speech-5-0-turbo-ctc-ivnplcd09
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: Extremely Fast and Accurate Transcription with Granite Speech 5.0 Turbo CTC | daily.dev
og:description: IBM released two new 470M-parameter English speech recognition models in the Granite Speech family: granite-speech-5.0-470m-turboctc (Apache 2.0) and...
og:url: https://daily.dev/posts/extremely-fast-and-accurate-transcription-with-granite-speech-5-0-turbo-ctc-ivnplcd09
og:image: https://api.daily.dev/og/posts/ivNPlcd09.png
og:image:alt: Extremely Fast and Accurate Transcription with Granite Speech 5.0 Turbo CTC
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Extremely Fast and Accurate Transcription with Granite Speech 5.0 Turbo CTC

**[Hugging Face](https://daily.dev/sources/huggingface)** · 6 min read · 1 upvotes · 0 comments

## Summary

IBM released two new 470M-parameter English speech recognition models in the Granite Speech family: granite-speech-5.0-470m-turboctc (Apache 2.0) and granite-speech-5.0-470m-turboctc-nc (CC-BY-NC-SA-4.0, trained on more data). Both are encoder-only Conformer-based models achieving over 12,600 RTFx throughput on an NVIDIA H200 GPU, meaning they can transcribe more than 3.5 hours of speech per second with batched inference. Aggregate word error rates on OpenASR Leaderboard public test sets are 4.85% for the noncommercial model and 5.00% for the Apache 2.0 model. Unlike prior Granite Speech models with an LM and LoRA adapters, these drop speech translation and keyword biasing but gain over 20x faster throughput and a smaller footprint, making them suited for edge speech-to-text. The models generate 12.5 tokens per second (down from 50 characters per second in earlier encoders) via three stages of 2x subsampling, and are trained on natural datasets (MLS, YODAS, Common Voice, LibriSpeech, VoxPopuli, AMI, GigaSpeech, SPGISpeech) plus synthetic data generated with gpt-oss models and StyleTTS2. The models are natively supported in a source install of transformers.

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://huggingface.co/blog/ibm-granite/granite-speech-5-0-470m-turboctc>

## Questions this post answers

### What is the transcription speed of IBM's granite-speech-5.0-470m-turboctc model on an H200 GPU?

It reaches over 12,600 RTFx on an NVIDIA H200 GPU, meaning it can transcribe more than 3.5 hours of speech in one second using batched inference. This is achieved with a compact 470M-parameter encoder-only architecture, over 20x faster than previous Granite Speech models that included an LM with LoRA adapters.

_Teams evaluating fast, low-footprint ASR models can track benchmarks like this via daily.dev._

### What is the difference between granite-speech-5.0-470m-turboctc and granite-speech-5.0-470m-turboctc-nc?

The turboctc model is Apache 2.0 licensed, trained on a smaller dataset, and scores 5.00% aggregate WER, while the turboctc-nc variant trains on additional data, uses CC-BY-NC-SA-4.0 licensing, and scores a slightly better 4.85% aggregate WER, with a notable edge on SPGI Speech but a disadvantage on the chunked Earnings22 test.

_Developers choosing between licensing and accuracy trade-offs can follow model comparisons on daily.dev._

### Why do the new Granite Speech 5.0 encoder models operate at 12.5 tokens per second instead of 50 characters per second like earlier Granite Speech encoders?

They use three stages of 2x subsampling to reduce the log Mel spectrogram front end's 100 frames per second down to 12.5 tokens per second, combining a reshape-based stacking stage with two strided-convolution stages built into the first two Conformer blocks, trading token granularity for much faster inference throughput.

_Engineers optimizing ASR architectures for speed can dig into encoder design details on daily.dev._

## Similar posts on daily.dev

- [Granite Speech 5.0: Transcribe, Translate, and Attribute more Accurately and Faster](https://daily.dev/posts/granite-speech-5-0-transcribe-translate-and-attribute-more-accurately-and-faster-nyvlfrpof) · Hugging Face · 0 upvotes · 0 comments
- [Granite 4.0 1B Speech: Compact, Multilingual, and Built for the Edge](https://daily.dev/posts/granite-4-0-1b-speech-compact-multilingual-and-built-for-the-edge-djlhogrk3) · Hugging Face · 22 upvotes · 0 comments
- [Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint](https://daily.dev/posts/granite-switch-4-1-chat-decisions-and-speech-from-one-endpoint-0ig1nmhow) · Hugging Face · 1 upvotes · 1 comments

---

Tags: [#deep-learning](https://daily.dev/tags/deep-learning), [#transformers](https://daily.dev/tags/transformers), [#speech-recognition](https://daily.dev/tags/speech-recognition), [#huggingface](https://daily.dev/tags/huggingface)

[View this post on daily.dev](https://daily.dev/posts/extremely-fast-and-accurate-transcription-with-granite-speech-5-0-turbo-ctc-ivnplcd09)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"Extremely Fast and Accurate Transcription with Granite Speech 5.0 Turbo CTC","url":"https://daily.dev/posts/extremely-fast-and-accurate-transcription-with-granite-speech-5-0-turbo-ctc-ivnplcd09","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/extremely-fast-and-accurate-transcription-with-granite-speech-5-0-turbo-ctc-ivnplcd09"},"datePublished":"2026-08-25T15:01:34.914Z","dateModified":"2026-09-13T20:47:17.825Z","description":"IBM released two new 470M-parameter English speech recognition models in the Granite Speech family: granite-speech-5.0-470m-turboctc (Apache 2.0) and...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/5f367066a69a600e4b58d8d680b2bc1d?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/5f367066a69a600e4b58d8d680b2bc1d?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Hugging Face","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Hugging Face","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/f1f55c67d81a4330acf5b90b26b0c8e1","url":"https://daily.dev/sources/huggingface"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/extremely-fast-and-accurate-transcription-with-granite-speech-5-0-turbo-ctc-ivnplcd09","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"deep-learning,transformers,speech-recognition,huggingface","timeRequired":"PT6M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hugging Face","item":"https://daily.dev/sources/huggingface"},{"@type":"ListItem","position":3,"name":"Extremely Fast and Accurate Transcription with Granite Speech 5.0 Turbo CTC"}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/extremely-fast-and-accurate-transcription-with-granite-speech-5-0-turbo-ctc-ivnplcd09#faq","mainEntity":[{"@type":"Question","name":"What is the transcription speed of IBM's granite-speech-5.0-470m-turboctc model on an H200 GPU?","acceptedAnswer":{"@type":"Answer","text":"It reaches over 12,600 RTFx on an NVIDIA H200 GPU, meaning it can transcribe more than 3.5 hours of speech in one second using batched inference. This is achieved with a compact 470M-parameter encoder-only architecture, over 20x faster than previous Granite Speech models that included an LM with LoRA adapters. Teams evaluating fast, low-footprint ASR models can track benchmarks like this via daily.dev."}},{"@type":"Question","name":"What is the difference between granite-speech-5.0-470m-turboctc and granite-speech-5.0-470m-turboctc-nc?","acceptedAnswer":{"@type":"Answer","text":"The turboctc model is Apache 2.0 licensed, trained on a smaller dataset, and scores 5.00% aggregate WER, while the turboctc-nc variant trains on additional data, uses CC-BY-NC-SA-4.0 licensing, and scores a slightly better 4.85% aggregate WER, with a notable edge on SPGI Speech but a disadvantage on the chunked Earnings22 test. Developers choosing between licensing and accuracy trade-offs can follow model comparisons on daily.dev."}},{"@type":"Question","name":"Why do the new Granite Speech 5.0 encoder models operate at 12.5 tokens per second instead of 50 characters per second like earlier Granite Speech encoders?","acceptedAnswer":{"@type":"Answer","text":"They use three stages of 2x subsampling to reduce the log Mel spectrogram front end's 100 frames per second down to 12.5 tokens per second, combining a reshape-based stacking stage with two strided-convolution stages built into the first two Conformer blocks, trading token granularity for much faster inference throughput. Engineers optimizing ASR architectures for speed can dig into encoder design details on daily.dev."}}]}
```

