Parakeet Batch Transcriber

Unlimited offline batch speech-to-text that works on medium-grade consumer hardware, with a respectable 93% transcription accuracy and diarization (aka "who said what").

What it is

Transcribe audio and video files in bulk — entirely on your own PC, no cloud, no subscription, no data leaving your machine. Unlimited and fairly quick too.

Just toss your audio/video files into a folder, run the exe, select said folder, select another folder where to drop the text files and let it process - fire & forget, fully automated.

What you get per file

• Timestamped transcript (.txt)
• Clean readable text, no timestamps (.txt)
• SRT and VTT subtitles — drop straight into a video editor or media player
• Full JSON with per-word timestamps for devs and pipelines

(See screenshots I took in notepad++ of what the results look like)

Pros/Cons

Best for: Interviews, meetings, podcasts, lectures, voiceovers, narration — clear speech at reasonable recording quality. 

Batch-processing large archives overnight. Any recording you can't send to a cloud service: therapy sessions, legal recordings, confidential calls, NDA-covered content.

Sucks for: Music lyrics, hypnotic/meditation sessions, heavy background noise, distorted or phone-quality audio, very strong accents, or sound-effect-only files. 

If you want higher quality and don't mind paying for it, check out ElevenLabs Audio Transcriber — a companion app using cloud-based transcription for harder cases and richer output.

System Requirements:

• Window 10/11
• 8GB of RAM
• 4 GB of GPU VRAM
• AMD Ryzen 5 or equivalent (3.0 GHz CPU or better)
• 1 GB of storage space
• FFMPEG installed to path

Note 1: The above are recommendations, it might run on lower hardware too. GPU gives 10–20× faster transcription; CPU-only works too, just slower. 

There is an option to do a one-time model download (~715 MB, including Parakeet V3 and SpeechBrain) via Hugging Face, but the provided zip package you'll download on this page has them already pre-included so you can just boot up the exe and do your transcriptions right away.

FFMPEG Required Too:

You will need FFMPEG installed to path for this program to work, get it here:

https://reactorcore.itch.io/ffmpeg-to-path-installer 

(FFMPEG is a program that edits audio/video, but works in command line, my program needs it to process the audio/video you want to transcribe.

Other features:

• Speaker identification — labels who's talking (SPEAKER_00, SPEAKER_01, etc.)
• Vocal enhancement — optional HPSS pre-processing to help recover lyrics buried in music
• 25 languages + auto-detect
• Skips already-transcribed files automatically
• Fully portable — copy the folder to a USB drive, works offline on any Windows PC
• Super easy to use even for laypeople,
• Intuitive visual GUI with tooltips and labels
• Has a log.txt, settings menu and various option to help you succeed

Unlimited, automated, good. 

No subscriptions. No data leaves your machine.

---

Technical Stuff

The quality is not bad - at around 93% accuracy - so as long as your audios are clean enough and spoken in clear english - and without background noise - you will get fairly good results with minimal word errors. Other languages or strong accents might have a higher Word Error Rate (WER). 

For something as small and lightweight as Parakeet V3 that stays under 1GB in size and doesn't require humongous VRAM usage, this is impressive - its the best unlimited consumer grade option you can get at the moment.

Again, if you need more quality/accuracy, audio event tagging and better speaker detection use Elevenlabs Audio Transcriber, but notice that it uses cloud service that requires an internet connection, an account and an API key.

---

Check out everything else I do: 🚀✨

https://linktr.ee/reactorcore