IBM released Granite 4.0 1B Speech — a compact multilingual speech model for edge AI

IBM released Granite 4.0 1B Speech — a compact speech-language model for multilingual speech recognition and bidirectional translation. What's important in this news is not just a new release, but IBM's bet on production scenarios where memory, latency, and inference cost are as critical as benchmark quality.

What changed

Granite 4.0 1B Speech replaces heavier configurations in the Granite Speech lineup and focuses on efficiency. According to IBM, the model has half the parameters of granite-speech-3.

3-2b, while achieving improved English ASR accuracy, support for Japanese speech recognition, keyword list biasing, and faster inference through encoder fine-tuning and speculative decoding. The idea is simple: don't increase size at any cost, but remove excess weight without losing core capabilities that teams need in real production. IBM emphasizes the training approach separately.

The model is built on granite-4.0-1b-base, which was fine-tuned on speech tasks through modality alignment. The training mix included open ASR and AST corpora, as well as synthetic datasets for Japanese language, keyword-biased ASR, and speech translation.

For developers, this is an important signal: IBM is not building a closed voice stack only for the cloud, but developing an open model that can be adapted to your own pipelines and hardware.

Languages and tasks

Granite 4.0 1B Speech is designed for enterprise scenarios where both transcription and bidirectional speech translation are needed. The basic set of supported input languages includes English, French, German, Spanish, Portuguese, and Japanese. For translation, IBM positions the model as a tool for speech-to-text and speech translation to English and from English for these languages, and separately specifies English-to-Italian and English-to-Mandarin scenarios. This makes the release useful not only for call centers and voice interfaces, but also for internal translation pipelines.

• Speech recognition in English, French, German, Spanish, Portuguese, and Japanese
• Bidirectional speech translation for pairs with English
• Separate English-to-Italian and English-to-Mandarin scenarios
• Biasing by keyword list for names, brands, and abbreviations
• Operation in scenarios where low latency and limited memory are critical

Another practical advantage is the Apache 2.0 license. For enterprise teams, this reduces friction at the pilot and legal review stage: the model can be deployed locally, embedded in your own stack, and not tied to API-only access at an early stage. Against a market where many speech systems are available only as a cloud service with commercial restrictions, this format gives more freedom for customization, offline deployment, and data control.

Deployment and metrics

According to the model card, Granite 4.0 1B Speech has already topped the OpenASR leaderboard with an average WER of 5.52 and RTFx of 280.

02. In the dataset breakdown, IBM shows, for example, 1.42 on LibriSpeech Clean, 2.

85 on LibriSpeech Other, and 3.10 on Tedlium. For such releases, this is an important argument: the model is positioned not just as "small and cheap," but as a compact system that still maintains a very strong level on standard public tests.

In terms of deployment, IBM tried to remove unnecessary barriers. The model is supported in **transformers 4.52.

1+, runs via vLLM, and has a separate path for mlx-audio** on Apple Silicon. The reference pipeline uses mono audio at 16 kHz, the request is formed via the `<|audio|>` prefix, and keyword biasing can be added directly to the prompt. Architecturally, Granite Speech remains a two-pass system: first the model converts audio to text, then if needed a separate language model call processes the transcript.

For production, this is convenient because recognition and downstream logic can be scaled and tuned independently.

What this means

IBM is betting on the voice AI segment where the winning model is not the largest one, but the one that can actually run on limited resources without losing quality. If Granite 4.0 1B Speech takes hold in production deployments, the market will get another strong open-source option for local transcription, speech translation, and edge services without heavy cloud dependence.