tUtils - Terminal Utilities

A collection of standalone scripts and utilities. All Python scripts use PEP 723 inline metadata — uv run installs dependencies automatically.

Contents

transcribe_audio_folder.py

Batch audio-to-SSML transcription using OpenAI Whisper large-v3. Recursively processes audio files in a folder and creates .ssml files alongside each source, compatible with generate_speech.py for re-synthesis.

• Supports 12+ audio formats (m4a, mp3, wav, flac, ogg, opus, etc.)
• Auto-downloads the Whisper large-v3 model (~3GB) to ~/Downloads/whisper_models/
• Speaker diarization via pyannote.audio — assigns distinct Chirp 3 HD voices per speaker (up to 6)
• Dublin Core metadata (<dc:title>) embedded in SSML output
• Skips already-transcribed files on re-runs

Diarization requires a HuggingFace token (HF_TOKEN env var) and accepted access on:

• huggingface.co/pyannote/speaker-diarization-3.1
• huggingface.co/pyannote/segmentation-3.0
• huggingface.co/pyannote/speaker-diarization-community-1

Google Cloud Media Generation Scripts

Script	Model	API
generate_video.py	Veo 3.1 (veo-3.1-generate-001)	Vertex AI
generate_audio.py	Lyria 002 (lyria-002)	REST
generate_image.py	Imagen 4.0 Ultra (imagen-4.0-ultra-generate-001)	google-genai SDK
generate_speech.py	Chirp 3 HD	Cloud Text-to-Speech Long Audio Synthesis

generate_video.py — Generates videos with image-to-video continuation (extracts last frame via ffmpeg for chaining). Outputs to video/ with video-series-{theme}-{timestamp}.mp4 naming.

generate_audio.py — Music generation with prompt, negative prompt, seed, and sample count parameters. Outputs to audio/.

generate_image.py — Reads prompt from ~/my/src/loxal/lox/al/prompts/avatar.md. Outputs to image/.

generate_speech.py — Reads SSML from ~/my/src/loxal/lox/al/prompts/speech.ssml (falls back to speech.md). Supports --lang flag for language/voice selection (de, en, ru). Extracts Dublin Core metadata from SSML and embeds it as ID3 tags in the output MP3. Uses GCS bucket for long audio staging, converts LINEAR16 WAV to MP3 via ffmpeg.

Google Cloud Authentication

1. Install the gcloud CLI.

2. Authenticate and set the quota project:

gcloud auth login
gcloud auth application-default login
gcloud auth application-default set-quota-project YOUR_PROJECT_ID

3. Enable the required APIs:

gcloud services enable aiplatform.googleapis.com --project=YOUR_PROJECT_ID
gcloud services enable texttospeech.googleapis.com --project=YOUR_PROJECT_ID

4. Ensure your account has the Vertex AI User role:

gcloud projects add-iam-policy-binding YOUR_PROJECT_ID \
  --member="user:YOUR_EMAIL" \
  --role="roles/aiplatform.user"

hurl-to-har-to-hurl-converter/

Bidirectional converter between HURL and HAR formats. See hurl-to-har-to-hurl-converter/README.md.

Setup

• Python 3.10+, uv, FFmpeg

brew install ffmpeg   # macOS

Running

All recipes are in the justfile:

just generate-video
just generate-audio
just generate-image
just generate-speech           # default: German
just generate-speech en        # English
just generate-speech ru        # Russian
just transcribe
just merge-videos