Understanding audio speed change
Faster, same pitch.
Why "speed up the podcast" doesn't have to make everyone sound like a chipmunk, and the WSOLA algorithm that lives in every audio player.
Speed without pitch shift — the trick.
Time-stretching audio without changing pitch needs the dual of pitch-shifting: stretch the time axis while keeping the frequency content the same. The WSOLA algorithm (Waveform Similarity Overlap-Add) cuts the signal into short windows, slides them closer together (faster) or further apart (slower) with overlap, and adjusts the overlap to keep the waveform continuous. The result is the same content rendered in a different duration with the same pitch.
1.5×, 2×, 3× — the typical settings.
Podcast listeners commonly play at 1.5× or 2×; serious podcast consumers reach for 2.5-3×. Speech remains intelligible up to roughly 3× for most listeners and 3.5× for trained ones. Above that, the artifacts (slightly metallic vowels, flattened consonants) start hurting comprehension. For music, the practical speedup ceiling is much lower — 1.2-1.5× is the threshold before listeners notice something off in the timing.
A worked speedup.
Speed up a 60-minute podcast to play in 40 minutes (1.5× speed). ffmpeg -i in.mp3 -af "atempo=1.5" out.mp3. The atempo filter implements WSOLA; result is a 40-minute file with the same pitch as the original. The audio sounds noticeably faster but every word is still clearly audible. For ratios outside 0.5-2.0, atempo has to be chained: atempo=2.0,atempo=1.5 for 3.0×.
1.5× tempo, same pitch
atempo=1.5
WSOLA preserves pitch by sliding overlapping windows.
ffmpeg -i in.mp3 -af atempo=1.5 out.mp3
= 60 min → 40 min, same voice pitch
Speed change vs pitch shift, both wrong.
To both shift pitch and change speed (rare, but the "vintage tape" effect uses it): just resample. ffmpeg -i in.wav -af "asetrate=66150,atempo=1.0" out.wav replays at 1.5× the sample rate, making the audio 1.5× faster AND raising pitch by 7 semitones. That's the "wrong RPM" effect. Combined effects need both tools composed together.
Listening efficiency math.
Reading speed is roughly 200-300 wpm; conversational speech is ~150 wpm; podcast dialogue is 150-180 wpm. Playing a podcast at 1.5× brings it to 225-270 wpm — close to reading speed. 2× pushes it to ~300-360 wpm, slightly faster than reading. The comprehension cost is small for technical content but real for narrative content that wants pauses to land. For audiobooks, 1.25-1.5× is the sweet spot for most listeners.