Glossary
Generative error correction (GER)
Generative error correction (GER) passes a speech recognizer's output, often its N-best list of alternative transcripts, to a large language model, which rewrites the transcript to fix errors. It can recover words the recognizer never offered, but it can also change words that were already right.
Updated October 17, 20264 cited facts, each traced to its source
Also called GER, LLM post-correction, ASR error correction.
Generative error correction (GER) is a step that takes the output of a speech recognizer and asks a large language model to write a corrected version. The term comes from a 2023 study that asked whether LLMs can act as post-processors for speech recognition, doing rescoring and error correction on a recognizer's output 1. The word "generative" is the key difference: the model writes new text rather than only choosing among the candidates the recognizer already produced.
How the step works
A recognizer does not return only one transcript. It usually produces an N-best list, a ranked set of alternative transcripts for the same audio. A GER step gives the LLM the top transcript, or the whole list, and asks for the most likely correct sentence. Because the output is written by the model, it can differ from every item on the list.
It can add words the list does not contain
Consider a recognizer that hears a name it has never seen. Its alternatives may all be wrong, and none may contain the right spelling. A 2023 study on the HyPoradise benchmark found that an LLM can correct words that are missing from the N-best list, because it generates text rather than only selecting among candidates 2. In this illustrative case, the model might produce a spelling that fits the sentence context, even though no alternative contained it.
It can also change words that were right
The same freedom causes a problem. A study of LLM correction for speech recognition reports that LLMs make many unnecessary changes to the input transcription, which the authors call over-correction 3. A 2025 INTERSPEECH paper says that text-only GER relies mainly on the text and ignores phonetic cues, which leads to over-correction 4. The authors' proposed fix is to give the model the recognizer's alternatives and phonetic context, so it can see which words sounded alike.
A concrete example
Suppose you dictate "send the file to Priya". The recognizer returns "send the file to pre-ya" as its top result. A GER step sees the sentence and a plausible name, and returns "send the file to Priya". That fix is what GER is for. Now suppose you dictate a technical term that is correct as heard. A GER step that prefers common words may swap it for a more familiar one, and you would not notice unless you reread the text.
Why it matters for dictation
GER is one of the places where an AI dictation tool can make your text better or worse. The raw recognizer keeps your words as heard, including errors. A GER step can fix names and phrasing but can also rewrite correct words without telling you. If you use a tool that rewrites dictation, check names, numbers and technical terms before you send the text. Our guide to AI cleanup for dictation explains the trade-offs, and speech to text without AI covers the option of raw output.
Related terms
- Large language model (LLM): the model that usually performs the correction.
- Language model (in speech recognition): the older statistical model that also helps a recognizer choose likely words.
- Word error rate (WER): the measure used to tell whether a correction step actually helped.
- Speech language model: a model that takes audio as input, which avoids the text-only step.
Guides that cover it
Evidence
Every number on this page, with where it comes from. Open a fact to see the exact words in its source.
- 1.
- 2.
- 3.
- 4.In the 2025 INTERSPEECH paper, existing language-model error correction for speech recognition relies mainly on the text, ignores phonetic cues, and that leads to over-correction, meaning the model changes words that were already right.
Natsuo Yamashita et al., arXiv (accepted by INTERSPEECH 2025), 2025 · Original
Related terms
- Speech to Text AI: Dictation, Assistants and Voice AgentsThe topic this term belongs to
- AI text detectorAn AI text detector is software that estimates whether a passage was written by a language model, usually from the words alone. Detectors make mistakes in both directions, so a score is weak evidence about how a text was made.
- Generative AIGenerative AI is the class of AI models that create new content, such as text, audio or images, from patterns in their training data. The NIST profile defines it as models that emulate the structure and characteristics of input data to generate derived synthetic content.
- Intent recognitionIntent recognition is the step that decides what a spoken request is asking for, such as setting a timer or sending a message, and which details the action needs. Voice assistants and voice agents rely on it before they act.
- Large language model (LLM)A large language model (LLM) is a language model with a very large number of parameters, trained on huge amounts of text to predict and generate language. Google's glossary defines the minimum as a language model having a very high number of parameters.
- Machine learningMachine learning is a method in which a program learns its behavior from example data rather than from rules written by hand. Google's glossary defines it as a program or system that trains a model from input data.
- Prompt (AI)In AI, a prompt is the input you give a model to get a response, such as a question, an instruction or pasted text. Google's glossary defines it as any text entered as input to a large language model to condition the model to behave in a certain way.
- Speech language modelA speech language model (speech LLM) is a large language model that accepts audio as input, so it can process speech directly instead of working only on a text transcript. The 2023 SALMONN model is one example that lets an LLM process general audio inputs.
- Speech-to-speech modelA speech-to-speech model takes audio in and produces audio out with one model, without writing the conversation out as text in between. Most voice systems instead chain speech recognition, a language model and speech synthesis.
- Text to speech (TTS)Text to speech (TTS), also called speech synthesis, is software that converts written text into spoken audio. Modern systems use neural networks, and they can run on the device or in the cloud.
- Turn detectionTurn detection is the step that decides when a speaker has finished a turn, so a voice system knows when to respond. A pause can be a hesitation rather than an ending, so the decision trades speed against cutting the speaker off.
- Voice agentA voice agent is a voice system that listens, reasons with a language model, uses tools, and speaks back across several turns to finish a task, such as checking an order and rescheduling it. Benchmarks show that agents still fail many tasks on realistic audio.
- Voice assistantA voice assistant is software that listens for a spoken request, works out what you want, and then acts or answers, such as setting a timer or turning off a light. Dictation is different: it turns your speech into text for you to edit.
Pages that use this term
- Speech to Text AI: Dictation, Assistants and Voice AgentsTopic
- Is Speech to Text AI? What Kind of AI Voice Typing UsesGuide
- AI Cleanup for Dictation: Fillers, Punctuation, RewritesGuide
- Speech to Text and LLMs: What Language Models ChangeGuide
- Do LLMs make speech recognition more accurate?Question
- Is speech to text generative AI?Question
- Is transcription software considered generative AI?Question
- What is the difference between speech to text and AI?Question
- AI text detectorGlossary
- Generative AIGlossary
- Large language model (LLM)Glossary
- Speech language modelGlossary