Kokoro TTS for YouTube Voiceovers: A Step-by-Step Python Guide on Mac

mkdir ~/youtube-voiceover && cd ~/youtube-voiceover
python3 -m venv venv
source venv/bin/activate

pip install kokoro soundfile

from kokoro import KPipeline
import soundfile as sf

pipeline = KPipeline(lang_code="a")
generator = pipeline("Welcome to today's video. Let me show you how this works.", voice="af_heart")

for i, (_, _, audio) in enumerate(generator):
    sf.write(f"test_{i}.wav", audio, 24000)

print("Done — check test_0.wav")

python test.py

Welcome back to the channel. Today we are breaking down the three tools every creator needs.

First up is screen recording software. You want something lightweight that captures at 60 frames per second.

Second is an audio tool. Bad audio kills a video faster than bad footage.

Third is a local text-to-speech engine so you can iterate on voiceovers without re-recording.

That wraps up this video. Drop a comment with your favorite tool and I will see you in the next one.

from kokoro import KPipeline
import soundfile as sf

with open("script.txt") as f:
    text = f.read()

pipeline = KPipeline(lang_code="a")
generator = pipeline(text, voice="af_heart")

for i, (_, _, audio) in enumerate(generator):
    sf.write(f"full_{i}.wav", audio, 24000)

print(f"Generated {i+1} file(s)")

voices = ["af_heart", "af_bella", "am_adam", "am_liam"]
pipeline = KPipeline(lang_code="a")

for v in voices:
    gen = pipeline("This is a voice preview for YouTube narration.", voice=v)
    for i, (_, _, audio) in enumerate(gen):
        sf.write(f"preview_{v}.wav", audio, 24000)

##hook
Ever feel like your videos take too long to produce? Here is the fix.

##body
Start with a template. Every video I make follows the same four-part structure: hook, setup, demonstration, and call to action.

The template saves about two hours per video because I am not staring at a blank timeline.

Next batch your voiceover. Generate all your narration at once, then edit. Do not generate line by line.

##outro
That is the workflow. Try it on your next video and let me know if it saves you time.

from kokoro import KPipeline
import soundfile as sf

with open("script_sections.txt") as f:
    raw = f.read()

sections = {}
current = None
for line in raw.split("\n"):
    if line.startswith("##"):
        current = line[2:].strip()
        sections[current] = []
    elif current and line.strip():
        sections[current].append(line.strip())

pipeline = KPipeline(lang_code="a")
voice = "af_heart"

for name, lines in sections.items():
    text = " ".join(lines)
    gen = pipeline(text, voice=voice)
    for i, (_, _, audio) in enumerate(gen):
        sf.write(f"{name}_{i}.wav", audio, 24000)
        print(f"Wrote {name}_{i}.wav")

print("All sections generated.")

brew install ffmpeg
pip install pydub

import os
from pydub import AudioSegment

for f in os.listdir("."):
    if f.endswith(".wav"):
        audio = AudioSegment.from_wav(f)
        mp3_name = f.replace(".wav", ".mp3")
        audio.export(mp3_name, format="mp3", bitrate="192k")
        print(f"Converted {f} → {mp3_name}")

chunks = [p.strip() for p in text.split("\n\n") if p.strip()]
pipeline = KPipeline(lang_code="a")

for idx, chunk in enumerate(chunks):
    gen = pipeline(chunk, voice="af_heart")
    for i, (_, _, audio) in enumerate(gen):
        sf.write(f"chunk_{idx:03d}_{i}.wav", audio, 24000)

import sys
from kokoro import KPipeline
from pydub import AudioSegment
import soundfile as sf

if len(sys.argv) < 2:
    print("Usage: python render.py script.txt [voice]")
    sys.exit(1)

script_path = sys.argv[1]
voice = sys.argv[2] if len(sys.argv) > 2 else "af_heart"

with open(script_path) as f:
    text = f.read()

# Section split on ## markers
sections = {}
current = None
for line in text.split("\n"):
    if line.startswith("##"):
        current = line[2:].strip()
        sections[current] = []
    elif current and line.strip():
        sections[current].append(line.strip())

pipeline = KPipeline(lang_code="a")
wav_files = []

for name, lines in sections.items():
    content = " ".join(lines)
    gen = pipeline(content, voice=voice)
    for i, (_, _, audio) in enumerate(gen):
        fname = f"{name}_{i}.wav"
        sf.write(fname, audio, 24000)
        wav_files.append(fname)
        print(f"  Created {fname}")

# Convert to MP3
print("\nConverting to MP3 ...")
for wav in wav_files:
    segment = AudioSegment.from_wav(wav)
    mp3 = wav.replace(".wav", ".mp3")
    segment.export(mp3, format="mp3", bitrate="192k")
    print(f"  {wav} → {mp3}")

print(f"\nDone. Created {len(wav_files)} audio file(s).")

python render.py script_sections.txt af_heart

Good. Now watch this.

Good, now watch this.

# Instead of "Kapwing"
text = text.replace("Kapwing", "Cap-wing")
# Instead of "RTX 4090"
text = text.replace("RTX 4090", "are tee ex 4090")
# Instead of "Canva"
text = text.replace("Canva", "Canva")

fixes = {
    "Kokoro": "Ko-ko-ro",
    "KPipeline": "K-Pipeline",
    "RTX": "are tee ex",
    "UI": "U I",
    "README": "read me",
}
for wrong, right in fixes.items():
    text = text.replace(wrong, right)

from pydub import AudioSegment

audio = AudioSegment.from_wav("hook_0.wav")
trimmed = audio.strip_silence(threshold=-50, padding=50)
trimmed.export("hook_0_trimmed.wav", format="wav")