Text to Speech for Online Course Narration

The Course Creator's Recording Problem

Creating an online course is already a massive effort. Writing the curriculum, building slides, recording screen captures, editing video. Then comes narration: sitting in a quiet room, recording 50+ lessons, re-recording when you stumble over a sentence, re-recording again when you update the content six months later.

For a typical 10-hour course, recording and editing narration takes 40 to 60 hours. That is before post-production (noise removal, volume normalization, breath editing). If you hire a professional narrator, expect to pay $2,000 to $4,000 for a 10-hour course. And every time you update a lesson, you either re-record it yourself or pay for another session.

AI text-to-speech changes this equation fundamentally. Write your script, generate audio, and export. Need to update lesson 37? Change the text and regenerate. The voice stays consistent, the cost stays zero (after the initial purchase), and the turnaround drops from hours to minutes.

Choosing a Voice for Educational Content

The right voice for course narration is one students can listen to for hours without fatigue. That means clear diction, moderate pace, and a warm but neutral tone. Avoid voices that are overly expressive or dramatic. Education content needs to inform, not perform.

In Murmur, Kokoro voices are the best starting point for course narration. The model maintains remarkably consistent tone and pacing across long sessions, which matters when students are listening to lesson after lesson. For courses that benefit from a more conversational feel (soft skills, creative topics), Qwen3 voices add warmth without sacrificing clarity.

Critical rule: use the same voice across every lesson in your course. Switching voices between modules is jarring and undermines the professional feel you are building. Pick one voice, test it with a sample lesson, and commit to it.

Batch Processing Workflow

For a full course, you will generate dozens of audio files. Here is the workflow that works:

1. Write all lesson scripts first. Do not generate audio until the scripts are reviewed and finalized. Regenerating is easy, but iterating on scripts while also managing audio files creates unnecessary complexity.
2. Name your scripts consistently: "01-introduction.txt", "02-getting-started.txt", etc. This keeps your audio exports organized.
3. Generate one test lesson and listen to the full thing. Check for mispronounced technical terms, awkward pauses, and pacing issues.
4. Adjust speed settings. For technical courses, 0.9x to 1.0x gives students time to absorb. For overview/conceptual lessons, 1.0x to 1.1x keeps energy up.
5. Generate all lessons with the same voice and speed settings. Consistency is non-negotiable.
6. Review each audio file. Regenerate any sections that sound off. With Murmur, there is no cost penalty for regeneration.

Handling Technical Terms

TTS models occasionally mispronounce specialized terminology, acronyms, or brand names. The fix is simple: spell it out phonetically in your script. Write "Kubernetes" as "koo-ber-net-eez" if the model stumbles. Write "SQL" as "sequel" if that is your preferred pronunciation. Most models handle common tech terms correctly, but always verify with a test generation.

Cost Comparison for a 10-Hour Course

Platform Requirements

Different course platforms have different audio requirements. Udemy requires audio at 48kHz sample rate with consistent volume levels. Teachable and Skillshare accept standard formats (MP3, WAV) without strict technical specifications. Murmur exports WAV at 44.1kHz or 48kHz, which meets or exceeds all major platform requirements.

For video lessons (which most platforms prefer), pair your TTS audio with screen recordings or slides. Import the audio track into your video editor (DaVinci Resolve, iMovie, or ScreenFlow) and sync it with your visuals. This workflow is significantly faster than recording voiceover live while screencasting.

Frequently Asked Questions