1
0
mirror of https://github.com/osmarks/random-stuff synced 2024-12-26 18:10:34 +00:00
random-stuff/lesswrong_to_tiktok.py
2024-02-25 14:26:53 +00:00

93 lines
2.7 KiB
Python

import wave, sys
import nltk
from balacoon_tts import TTS
from collections import namedtuple
import struct
from PIL import Image, ImageDraw
import cv2
import numpy
import subprocess
WIDTH = 400
AUDIO = "/tmp/x.wav"
VIDEO = "/tmp/x.avi"
BACKDROP = "/tmp/x.mkv"
OUTPUT = "/tmp/x.mp4"
def render_text(text: str):
render_params = {"font_size": 24}
im = Image.new("RGBA", (WIDTH, WIDTH))
draw = ImageDraw.Draw(im)
loc = [0, 0]
toks = text.split()
toks.reverse()
text_commands = []
while toks:
chunk = []
while draw.textbbox(loc, " ".join(chunk), **render_params)[2] < WIDTH:
if not toks: break
chunk.append(toks.pop())
else: toks.append(chunk.pop())
bbox = draw.textbbox(loc, " ".join(chunk), **render_params)
text_commands.append((tuple(loc), " ".join(chunk)))
loc[1] = bbox[3]
draw.rectangle([0, 0, WIDTH, loc[1]], fill="white")
for loc, text in text_commands:
draw.text(loc, text, fill="black", **render_params)
return im
Pause = namedtuple("Pause", ["length"])
text = open("/home/osmarks/Downloads/seq1.txt").read()
tts = TTS("/home/osmarks/Downloads/en_us_hifi_jets_cpu.addon")
#supported_speakers = tts.get_speakers()
speaker = "6670"
def chunks(text: str) -> list[str | Pause]:
out = []
for line in text.splitlines():
if line:
for sent in nltk.sent_tokenize(line):
out.append(sent)
out.append(Pause(0.5))
out.append(Pause(1))
return out
RATE = tts.get_sampling_rate()
FPS = 30
def wavblank(seconds):
return struct.pack(">h", 0) * round(seconds * RATE / 2)
blank_frame = render_text("")
fourcc = cv2.VideoWriter_fourcc(*"MJPG")
video_writer = cv2.VideoWriter(VIDEO, fourcc, FPS, (WIDTH, WIDTH))
total_dur = 0
with wave.open(AUDIO, "w") as fp:
fp.setparams((1, 2, RATE, 0, "NONE", "NONE"))
for chunk in chunks(text):
if isinstance(chunk, str):
samples = tts.synthesize(chunk, speaker)
image = render_text(chunk)
elif isinstance(chunk, Pause):
samples = wavblank(chunk.length)
image = blank_frame
fp.writeframes(samples)
duration = len(samples) / RATE # what
total_dur += duration
frame = cv2.cvtColor(numpy.array(image), cv2.COLOR_RGBA2BGR)
for _ in range(round(duration * FPS)): video_writer.write(frame)
print(chunk, duration)
video_writer.release()
subprocess.run([
"ffmpeg",
"-i", BACKDROP, "-i", AUDIO, "-i", VIDEO,
"-filter_complex", "overlay=x=200:y=200,format=nv12,hwupload",
"-to", str(total_dur),
"-y",
"-vaapi_device", "/dev/dri/renderD128", "-c:v", "h264_vaapi",
OUTPUT
]).check_returncode()