andras/perfect-postcode
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages 2 Wiki Activity Actions

migrate to chatterbox

Browse source
This commit is contained in:
Andras Schmelczer 2026-07-14 16:59:48 +01:00
parent 716e42a57d
commit 988c01c4f7
7 changed files with 480 additions and 1786 deletions
Download patch file Download diff file Expand all files Collapse all files

4
video/tts/mux.py
View file

@ -3,7 +3,7 @@
Reads two manifests inside ``output/<storyboard>/``:
* ``audio/index.json`` (synth output): per-cue WAV filename + measured
duration. Generated BEFORE recording in one batched Qwen3-TTS call.
duration. Generated BEFORE recording by ``tts/synth.py``.
* ``narration.json`` (recorder output): per-cue ``videoTimeMs`` against
the trimmed video. Generated DURING recording.
@ -30,7 +30,7 @@ from pathlib import Path
# Mixing the narration WAVs raw left the muxed track at ~-24 LUFS: viewers
# had to crank the volume, which reads as low production quality. Target a
# touch under the platform norm so speech-only audio never pumps.
LOUDNORM_TARGET = "I=-15:TP=-1.5:LRA=11"
LOUDNORM_TARGET = "I=-14:TP=-1.5:LRA=11"
def measure_loudness(cmd_head: list[str], filter_complex: str) -> dict[str, str] | None: