VoiceBlogger

Turn any voice recording into a polished blog post and Instagram captions — entirely offline, using local AI. No API keys, no cloud, no cost per run. Just free, private, and local goodness.

Available on Apple Silicon (M-series Mac) and as a native iOS app. The transcription layer uses mlx-whisper, which requires MLX and will not run on Intel Macs, Linux, or Windows.

(Don't worry, other platforms are coming sometime in the future)

---

How it works

1. Transcribe — mlx-whisper (Whisper large-v3) converts your audio to text. Pass --task translate to get English output from any language, or --task transcribe to keep the original language.
2. Polish — an Ollama LLM cleans up grammar and formats the transcript as a readable blog post.
3. Instagram captions — the same model generates 3–5 ready-to-post captions with hashtags.

All three outputs are saved as files next to your audio.

iOS app

The iOS app is a SwiftUI companion to the Mac workflow. On first launch it downloads the required on-device models once, then runs fully offline.

What it includes:

• One-tap recording with live waveform feedback
• Automatic transcription after you stop recording
• Blog generation and Instagram caption views
• A history screen for past posts stored locally on device
• Built-in model download and retry handling

The app lives under iOS App/VoiceBlogger.

---

Prerequisites

1. Apple Silicon Mac

MLX only runs on M-series chips (M1/M2/M3/M4).

2. Python dependencies

pip install -r requirements.txt

On first run, mlx-whisper will automatically download the Whisper large-v3 model (~3 GB).

3. Ollama

Install Ollama and pull a model:

# Install Ollama, then:
ollama pull gemma4:e4b   # default model (~9.6 GB)

# Or use any other model you have installed:
ollama list

Ollama must be running before you use this tool (ollama serve or the Ollama desktop app).

iOS app setup

The iOS app manages its own on-device model downloads on first launch, so there is no separate API-key or cloud setup. Once the models are present, the app works offline for recording, transcription, blog generation, and caption creation.

---

Usage

python Transcribe_and_blog.py <audio_file> [options]

Options

Flag	Default	Description
--language	hi	Source language (ISO 639-1 code, e.g. en, hi, es, fr)
--task	translate	translate → English output; transcribe → keeps source language
--model	gemma4:e4b	Ollama model name (run ollama list to see what you have)
--no-instagram	off	Skip Instagram caption generation

Examples

# Hindi audio → English blog + Instagram captions (defaults)
python Transcribe_and_blog.py my_recording.m4a

# English audio → English blog, no Instagram captions
python Transcribe_and_blog.py podcast.m4a --language en --task transcribe --no-instagram

# Spanish audio → English blog using a different model
python Transcribe_and_blog.py entrevista.mp3 --language es --task translate --model qwen2.5:14b

Project layout

voiceblogger/
├── Raw_Data/               # drop audio files here; raw transcripts saved here too
│   └── recording.m4a
├── blog/                   # polished blog posts (auto-created)
│   └── recording_blog.md
├── insta/                  # Instagram captions (auto-created)
│   └── recording_instagram.md
├── iOS App/VoiceBlogger/   # native SwiftUI iOS app
└── Transcribe_and_blog.py

Output files

Given Raw_Data/recording.m4a, three files are produced:

File	Contents
Raw_Data/recording_raw.txt	Raw Whisper transcript
blog/recording_blog.md	Polished blog post (Markdown)
insta/recording_instagram.md	Instagram captions with hashtags

---

Supported audio formats

Any format supported by Whisper: .m4a, .mp3, .wav, .mp4, .ogg, .flac, and more.

---

License

Apache License 2.0 — see LICENSE.