GPT Transcribe

GPT Transcribe is an online speech-to-text service that converts audio into written text using two models: gpt-transcribe for finished files and gpt-live-transcribe for low-latency live audio. It supports multilingual transcription and allows users to upload or record audio in formats such as mp3, mp4, m4a, wav, ogg, and webm. Users can optionally add scene prompts, keywords, and language hints to guide the transcription process.

The platform offers a free tier with daily check-in credits, a Pro plan at $19.99 per month, and a Max plan at $49.99 per month. All plans provide access to the online generator where users can run jobs, edit transcripts, and copy text to their clipboard. The Pro plan includes up to ~5,328 total audio minutes per year, while the Max plan offers up to ~13,328 total audio minutes per year.

Users can upload or record audio in file mode for finished recordings or live mode to capture microphone audio with interim captions. The process involves bringing the audio, adding optional context like scene prompts and keywords, running the transcription, and then copying and refining the result inline. Live sessions stream interim text before finalizing the capture for a cleaner transcript.

Some reviewers on Reddit say the service understands context well, even when users speak gibberish, and accordingly fixes words in the transcript. Others note that it performs better than Whisper on accents, short phrases, numbers, and specialized terms, with a real-world word error rate of 8.98% compared to Whisper-1's 15.21%.

Some users on Reddit mention that the service struggles with bilingual dictation where languages switch mid-sentence without clear separation. They also note that while it is good for many use cases, it may not be the best choice when speaker diarization is a strict requirement alongside realtime streaming.

The service supports multilingual transcription and allows users to add language hints to guide the model. It also provides an online generator where users can run jobs, edit the resulting transcripts, and copy them to their clipboard for further refinement.

The website states that gpt-transcribe is optimized for completed audio files and batch workloads like meetings, podcasts, and support recordings. In contrast, gpt-live-transcribe is built for low-latency live transcription over WebSocket / WebRTC for captions, calls, and streaming speech that is still being produced.

The platform allows users to add optional scene prompts, keyword lists, and language hints to help the model understand the audio. These context fields can be skipped if not needed, and users can re-run the transcription with better keywords if a critical term was missed.

The service offers a free tier with up to ~10 total audio minutes and a Pro plan with up to ~5,328 total audio minutes per year. The Max plan provides up to ~13,328 total audio minutes per year and is described as the strongest value for teams expecting heavy audio generation.

The website mentions that gpt-transcribe replaces whisper-1 as the recommended file STT starting point, achieving a real-world recording WER of 8.98% versus whisper-1's 15.21%. It is also noted as a successor recommendation to gpt-realtime-whisper for live STT.

Some Reddit users ask about the best path for production requirements involving realtime streaming, low latency partials, and speaker diarization. They also inquire about model cost comparisons for systems that listen to phone conversations and alert on certain phrases in close to realtime.

The platform supports audio formats such as mp3, mp4, m4a, wav, ogg, and webm with a maximum file size of 10MB. It also offers a free tier with daily check-in credits and allows users to upgrade anytime for more minutes.

The service is described as suitable for creators with ongoing audio needs, teams expecting heavy audio generation, and individuals looking for bilingual dictation support. It is particularly useful for meetings, calls, streams, and podcasts where context can be added to improve accuracy.

You can reach GPT Transcribe by e-mail at contact@gpt-transcribe.com.

The official website of GPT Transcribe is gpt-transcribe.com.