Quick answer: use Process → Time → Time Stretch or élastique Timestretch when duration must change while pitch stays fixed. Use Effects → Pitch → Shift when pitch must change while duration stays fixed. Use Pitch Bend for a change that moves over time. Use a tape-speed or varispeed process only when pitch and duration should move together.
The menu choice is the easy part. The decision that prevents most bad results is defining what must stay unchanged before processing. “Make it faster” can mean shorten a voice-over without raising the speaker’s voice, tune a flat note without moving the edit point, or reproduce tape running at the wrong speed. Those jobs need different processing and different quality checks.
Time Stretch vs Pitch Shift vs Varispeed
| Job | Duration | Pitch | Best starting point |
|---|---|---|---|
| Fit dialogue, a loop, or a cue to time | Changes | Preserved | Time Stretch or élastique Timestretch |
| Retune a voice, note, or sound effect | Preserved | Changes | Pitch Shift |
| Create a rise, fall, dive, or tape-stop shape | Usually preserved | Changes over time | Pitch Bend |
| Correct or imitate tape speed | Changes | Changes | Linked speed/pitch processing |
Calculate the Target Before Processing
Do not stretch by ear when the destination has an exact duration. Write down the original length and target length first. For a clip that must move from 30.000 seconds to 27.500 seconds, the target duration is 27.5 ÷ 30 = 91.67% of the original. Check which value the active dialog requests—new length, percentage, or ratio—before entering the number.
For tempo matching, use new duration = old duration × old BPM ÷ new BPM. An eight-second phrase recorded at 120 BPM must become 7.5 seconds to fit the same musical span at 128 BPM. That is a 93.75% duration target. This calculation avoids repeated preview-and-undo passes and gives you a result that lands on the intended edit point.
Leave a clean source untouched. If the clip needs trimming, do that non-destructively in a copy or prepare the boundaries with the Sound Forge cut and trim workflow. Every new algorithm comparison should start from the same original, not from a file that has already been stretched once.
Change Duration Without Changing Pitch
- Duplicate the file or save a working copy.
- Select only the passage that must change. With no selection, Sound Forge processes the entire file.
- Choose Process → Time → Time Stretch or Process → Time → élastique Timestretch.
- Enter the calculated target length or ratio.
- Choose the available mode that best matches speech, music, or the source material.
- Preview the hardest transient and the most exposed sustained sound, not just the quiet opening.
- Apply once, then verify the new duration, peaks, and both edit boundaries.
The current Sound Forge 2026 Time Stretch help documents both commands under the Process → Time submenu. The help does not promise that one algorithm is transparent on every source. That judgment still has to come from the material.
Which Source Exposes Which Problem?
Speech reveals doubled breaths, metallic consonants, and unstable vowel tone. Drums reveal softened attacks, pre-echo, and flams. Bass reveals a wandering note center or a blurred start. Sustained pads and ambience reveal chorusing, pulsing, and repeating texture. A setting that survives a drum loop may still fail badly on a solo voice.
A useful community signal is that there is no universal “best” stretcher. In an audio-engineering discussion comparing pitch-shift and time-stretch algorithms, engineers repeatedly recommend testing the actual source because percussion, bass, speech, solo instruments, and full mixes expose different artifacts. Treat that as directional experience, not as a benchmark for Sound Forge.
Choose the Method by Material
Start with the smallest change that solves the timing problem. A short voice-over correction often works better when you tighten natural pauses first and stretch only the remaining difference. A loop may need its first transient aligned before any algorithm is blamed. A long music master may sound more natural if the timing problem is repaired in the multitrack session rather than forced into one stereo file.
- Dialogue: preserve consonants and breaths; test the quietest exposed sentence.
- Drums: protect the leading edge; compare kick and snare attacks at matched level.
- Full mixes: inspect cymbals, reverb tails, and stereo image movement.
- Ambience: listen for a repeating cycle that was not present in the source.
- Loops: verify bar length and the seam after the loop repeats, not only the first play.
If timing relies on exact boundaries, add markers before processing and recheck them afterward. The markers and regions guide covers named reference points and region boundaries that make before/after timing easier to verify.
Protect Partial-Selection Boundaries
Processing the middle of a file creates two joins: untouched audio into the stretched selection, then the processed selection back into untouched audio. Even when the algorithm sounds clean, those joins can click, change ambience abruptly, or move a breath unnaturally. Place the selection boundary in stable room tone or between musical attacks when possible. Keep handles on both sides so the transition can be repaired without reopening the stretch process.
After rendering, audition from several seconds before the first boundary to several seconds after the second. A click heard only at the join is a boundary problem, not proof that the time-stretch algorithm failed. Repair the boundary with a short fade or crossfade from the original handles. Don’t hide a smeared transient under a long crossfade; move the boundary away from the transient or return to the original and use a better selection.
When the processed selection becomes shorter, everything after it moves earlier in a destructive single-file edit. That may be correct for a finished voice-over, but wrong for material tied to later markers, picture sync, or chapter times. Record the original and target positions before processing. For timing-sensitive work, render the changed section to a copy and replace it only after its start, end, and downstream markers have been checked.
Change Pitch Without Changing Duration
- Select the note, phrase, or complete file that needs transposition.
- Open Effects → Pitch → Shift.
- Enter the change in semitones or cents.
- Enable the mode that preserves duration when the edit point must not move.
- Preview the most exposed note and the transition into unprocessed audio.
- Apply from the untouched source and check peaks, tuning, and timbre.
The official Pitch help distinguishes Shift from Bend: Shift applies a fixed change, while Bend uses an envelope that moves pitch over time. Don’t confuse pitch shift with sample-rate conversion. Changing sample rate is a format and playback-rate issue; the sample-rate guide covers that separate workflow.
Semitones, Cents, and Formants
Twelve semitones equal one octave, and 100 cents equal one semitone. Cents are useful for small tuning corrections; semitones are easier for musical transposition. A technically correct +3-semitone shift can still make a voice sound smaller or cartoonish because vocal identity depends on resonances called formants, not only the fundamental pitch.
When a voice must remain natural, compare consonants, vowels, and breaths separately. If a global shift fixes one syllable but damages the rest, process a tighter selection or move to a note-aware editor. Avoid claiming that “formant preserve” means identity preserve—large moves still change articulation and texture.
Use Pitch Bend for Motion
Choose Effects → Pitch → Bend when the pitch must rise, fall, dive, or return over time. Sound Forge lets you draw an envelope, set its range in semitones, create up to 16 points, and choose curve types between points. The waveform can be shown in the graph, although the current help notes that it is unavailable when the selection exceeds 300,000 samples.
The Preserve original duration option keeps the edit length fixed. Boris FX says it works best for small corrections up to about ±2 semitones. That is useful guidance, not a cliff: difficult speech may reveal artifacts earlier, while a deliberately synthetic sound may tolerate much more.
When Varispeed Is the Correct Answer
Old tape, a turntable recorded at the wrong speed, and deliberate tape-stop effects are linked pitch-and-time problems. Correcting only duration can leave the material in the wrong key; correcting only pitch can leave it at the wrong speed. In these cases, the natural relationship is useful: speeding playback raises pitch and shortens duration, while slowing playback lowers pitch and lengthens it.
Before correcting an archival transfer, confirm that the error is constant. A machine that drifted during playback needs a changing correction, not one global percentage. Keep the original capture, document the applied ratio, and avoid normalizing until speed and pitch are settled.
Know When Not to Stretch
A large timing mismatch is often an editing problem disguised as a processing problem. Tighten pauses in narration before forcing every syllable through a heavy stretch. Recut a music cue on a bar or phrase boundary before compressing the entire mix. If the source is a multitrack session, repair timing at the track or stem level instead of asking one stereo master to preserve every transient and reverb tail at once.
Set a practical rejection point before listening. If the processed voice no longer sounds like the same speaker, the groove changes, or ambience develops an audible cycle, the correct answer is not another pass. Return to the source and change the edit, target, or tool.
When Melodyne or SpectraLayers Is Better
A global Sound Forge process moves everything in the selection. If only one note, syllable, or source inside a mixture needs correction, Melodyne or SpectraLayers may be more appropriate. Sound Forge supports ARA integration, but the external software must be installed and licensed for the workflow available on that machine.
The current ARA integration documentation gives two details worth following: the VST3 path must be available in Sound Forge’s VST Effects preferences, and Melodyne should be opened through Tools → Edit in Melodyne (ARA) rather than loaded into the ordinary Plug-In Chain. SpectraLayers has a corresponding Tools → Edit in SpectraLayers (ARA) command.
If a plugin does not appear, work through the Sound Forge VST troubleshooting guide before changing the audio or reinstalling unrelated components.
Final Quality Control
- Match playback level before comparing original and processed files.
- Play several seconds before and after both boundaries.
- Check the exact target duration or edit marker.
- Inspect new peaks; resynthesis can create sample values above the original peak.
- Listen in mono for unstable phase or image collapse.
- Compare speech consonants, drum attacks, bass note starts, and reverb tails.
- Save the processed master once, then perform format conversion from that master.
Do not normalize before deciding whether the process is acceptable; louder wins every blind comparison, which is exactly why I keep the levels matched until the stretch has been judged. Once the edit passes, use the normalization guide only if the delivery actually requires a peak or loudness adjustment. For file format changes, follow the audio conversion workflow instead of repeatedly resaving the stretched file.
Common Failures and Fixes
- The result lands at the wrong time: recalculate from original and target duration; do not guess the percentage.
- Speech sounds metallic: reduce the amount, try the alternate time-stretch method, and test a tighter selection.
- Drums have a soft double attack: return to the source and prioritize transient preservation.
- The voice is in tune but sounds like a different person: the pitch changed while the formant character did not survive naturally.
- A loop clicks after stretching: check the seam and apply a controlled boundary repair rather than stretching the processed result again.
- ARA opens like a normal plugin: use the Tools-menu ARA command instead of loading the plugin through the ordinary chain.
Time Stretch and Pitch Shift FAQ
Can Sound Forge change tempo without changing pitch?
Yes. Select the audio and use Process → Time → Time Stretch or élastique Timestretch. Enter the target duration, preview artifacts, and verify the final timing from the original source.
Can Sound Forge change pitch without changing length?
Yes. Use Effects → Pitch → Shift and preserve duration. Use cents for fine tuning and semitones for musical transposition, then check formants and edit boundaries.
Why does stretched audio sound metallic?
The algorithm is resynthesizing audio beyond what the source tolerates cleanly. Return to the untouched file, reduce the change, compare the alternate stretcher, and audition exposed consonants or sustained notes.
Should I stretch a file in several small passes?
No as a default workflow. Reprocessing an already stretched file accumulates damage. Compare settings as separate one-pass renders from the same original and keep the least damaged result.
Does Sound Forge include Melodyne or SpectraLayers?
Do not assume it does. Sound Forge supports ARA workflows, but installation, licensing, and edition availability depend on the products currently installed. Check the live vendor listing for your edition.
The Practical Rule
Define what must stay fixed, calculate the target before opening the process, render every comparison from the untouched source, and judge the hardest transient or exposed note before processing the full file.
Last fact-checked August 5, 2026 against the current Sound Forge 2026 online help.