Trim Audio Without Cutting Words

Open the Audio Trimmer & Fade Editor to select a continuous passage from local audio. This is a manual workflow, without AI transcription, automatic word detection, or speech-aware trimming.

Choose a complete thought before choosing a time

An outdoor instructor has supplied a voice clip explaining a route marker to a group. The useful sentence is “Before you follow the blue marker, pause at this sign.” An earlier sentence introduces the location, and a later sentence moves on to equipment. The editing task is to keep this particular direction understandable as a video voiceover. Start by listening to the whole passage and deciding which thought must survive. A cleanly cut word is not enough if its context disappears. Keeping “follow the blue marker” while removing the opening condition changes the instruction, even when every retained syllable sounds smooth.

Write the intended wording in your own review notes after listening, then compare it with the actual recording. The tool does not generate that wording or confirm it against a transcript. Include any small word that changes the relationship between actions, such as “before,” “after,” or “not.” If the speaker corrects a statement immediately afterward, decide whether the correction belongs in the selected thought rather than cutting solely at a convenient pause. The right range may be slightly longer than the most compact excerpt. Coherence matters more than reaching a neat duration with a fragment that now implies something different.

Separate the voice edit from its eventual presentation. The WAV will be an asset in a video editor or a permitted separate audio handoff, not a universal native audio attachment for every social platform. The video may need a view of the blue marker and then the sign, with enough time to connect the spoken references. Those pictures are not present in this tool. You choose the source, the continuous range, and the meaning to retain. There is no hidden recognition engine that locates sentences, removes filler, evaluates the instruction, or guarantees that a chosen boundary falls between complete words.

Start with audio that contains the missing edges

Use a local audio-only file with exactly one audio track, no video, and one or two channels. The source must be within 32 MiB and 180 seconds, with an original sample rate from 8000 to 192000 Hz. WAV, MP3, M4A or MP4 audio-only, OGG, FLAC, and audio WebM are browser-decoder dependent. If the instructor’s voice exists only inside camera footage, prepare an audio-only source elsewhere first. The trimmer does not extract video sound, fetch a link, record a microphone, or download a platform clip. A supported file extension cannot guarantee that its particular audio encoding will decode.

Listen for the opening breath and the last word in the source itself. If recording began halfway through “Before,” moving Start time to zero cannot recover the missing sound. If wind covers the final word, extending End time only retains more of that same masked signal. Find a fuller take or record another version outside the tool when the source is incomplete. This editor has no denoising, voice isolation, or reconstruction. A good source should include the whole utterance and some usable material around it. Spare room gives you choices; a recording already cut against the syllables gives you fewer.

The accepted source is prepared at 48000 Hz, potentially resampling its original rate while retaining mono or stereo. The initial range covers the first 30 seconds, or the entire prepared source when shorter. It is a default, not a detected sentence. Replacing the source with a new valid file resets settings except platform after confirmation; rejection or cancellation keeps the previous valid work. Keep the original recording separately and check which take is loaded. A remembered time from another take can point into the middle of a word even when both recordings have similar openings and the same overall duration.

Use the waveform as a map, not a word boundary

A large waveform rise often draws the eye, but the first meaningful sound can precede it. Some consonants are quiet, and the following vowel can be much more visible. In the outdoor recording, the beginning of “Before” may include a soft onset followed by a stronger vowel. Cutting at the strongest rise can remove part of the first syllable. Listen to a broader passage first, then enter an earlier Start time and prepare a candidate. The displayed bins come from actual decoded PCM samples, but their visual shape is not a language label or a reliable indication of where a word starts.

Leave pre-roll: a short retained lead before the word rather than an immediate cut onto it. The useful amount depends on the take, background sound, and delivery. Too little can make the voice seem switched on mid-syllable. Too much can preserve irrelevant walking noise or a fragment of the preceding sentence. There is no single millisecond value that solves both problems. Compare the candidate with the original at a consistent listening level and move the boundary deliberately. A little natural ambience can make the entrance easier to hear, provided it does not bring an unwanted earlier phrase into the asset.

Apply the same reasoning at the end. The visible activity can shrink before a final consonant, breath, or room decay has finished. Listen past the apparent end of the loud speech before choosing End time. A small tail can preserve the release of “sign” without retaining the next equipment sentence. Do not assume every low-amplitude area is silence that should be deleted. It might contain a quiet word or a meaningful breath. This tool does not offer automatic silence removal or word selection by clicking transcript text; the waveform is an orientation aid for the times you enter manually.

Connect the requested times to the selected sample frames

Suppose listening places the first syllable near 2.440 seconds and the final release near 18.720 seconds. Try Start time 2.310 and End time 18.950. The candidate lasts 16.640 seconds, with 0.130 seconds before the opening and about 0.230 seconds after the observed ending. These are hypothetical source observations, not detections produced by the tool. Prepare and listen before adopting them for a real take. If the prepared opening still feels truncated, move the start earlier and compare again. Entering a more precise number is not useful unless that number selects the sound you actually need.

The controls accept millisecond precision, while the 48000 Hz prepared source uses sample frames. Start follows ceil(start * 48000). An explicit end follows min(totalFrames, ceil(end * 48000)), and that end frame is excluded. For the example, start frame 110880 and end frame 909600 retain 798720 frames. Divide that count by 48000 to recover 16.640 seconds. This describes the mechanical selection exactly. It does not distinguish a consonant from background wind or prove the chosen excerpt preserves the intended instruction. Sample-frame consistency and a good spoken boundary are different things that both deserve a check.

A blank End time uses all remaining prepared source frames, not the next sentence boundary. The actual selected duration must be at least 0.25 seconds and no more than 120 seconds, within the loaded source. If times are reversed or outside the usable range, correct them rather than expecting a quiet automatic adjustment. The tool does not clamp an invalid selection to a nearby word. For a long source with the desired thought near the end, use explicit times and confirm the measured result after preparation. Source duration and export duration are separate bounds, not an invitation to retain everything available.

Keep breathing room outside the fade envelope

After the boundaries sound complete, consider whether the edges need any fade at all. In the example, a 0.060 second Fade-in would finish before the observed first syllable, leaving some lead at full level. A 0.150 second Fade-out can sit inside the tail if the final release really ends where you heard it. Both are candidates for listening, not automatic recommendations. A fade changes existing samples; it cannot insert pre-roll or restore a missing consonant. If the fade makes “Before” less distinct, shorten it or move the start earlier instead of increasing its length to make the entrance look smoother.

Fade-in and Fade-out are linear, accept zero to 10 seconds each, and cannot overlap. Their combined duration must not exceed the actual clip. Fade lengths are rounded to sample frames, and the first or last affected sample reaches zero at the corresponding edge. Zero leaves that edge unfaded. These are not crossfades between sentences, automatic breath edits, or adaptive volume envelopes. A long fade across speech may sound elegant in isolation while obscuring an important condition in the instruction. Judge the words at the same playback level with and without the fade before treating an even-looking waveform ramp as an improvement.

Keep a natural breath when it belongs to the delivery. Removing the intake immediately before a sentence can make the speaker sound rushed, while retaining an unrelated gasp from the previous take can be distracting. The useful boundary depends on what was actually recorded. If a breath sits inside the chosen range, the trimmer cannot remove only that internal event without also changing the continuous selection. Use another editor or a better supplied take for internal surgery. Here, adjust the outer edges and fades conservatively, preserving the first syllable, the sentence rhythm, and the completed last word in one coherent asset.

Prepare the WAV and listen without reading along

Begin with Level (dB) at 0 so the boundary comparison is not confused by a large volume change. Level accepts -24 to +12 dB in 0.1 dB steps and applies a fixed gain to the whole selection. It raises background noise along with speech and does not normalize loudness, denoise, or even out individual words. Output above full scale is refused rather than silently clipped; lower the gain and prepare again if necessary. Turning down an already clipped recording cannot reconstruct its damaged peaks. Solve a source problem with a better source instead of treating export acceptance as proof of repair.

Press Prepare audio, then play the prepared WAV from its beginning without reading your wording note. Can you hear the whole opening condition and the final reference to the sign? Knowing the sentence in advance can make a damaged syllable seem easier to understand than it really is. Ask another listener a neutral question about what they heard when practical. Compare with the original player, which includes the surrounding material. Neither player autoplays. Any input edit invalidates the previous result and dependent exports until preparation succeeds again, so test the rebuilt candidate rather than relying on a review of an older setting.

Use actual sample peak and RMS measurements as supporting information, not as word-completeness tests. They come from the exported WAV after re-decoding and are not LUFS, true peak, an intelligibility score, or platform certification. A quiet consonant can be missing without creating an obviously unusual number. A longer silent tail can change RMS while leaving the sentence unchanged. Silence may produce an unknown or null decibel reading. Keep the listening verdict specific: opening syllable intact, condition preserved, last word complete. Those observations say more about this task than a general declaration that the waveform or meters look correct.

Save the sound separately from its repeatable settings

Download the prepared audio as PCM16 WAV at 48000 Hz with mono or stereo retained. There is no MP3 or AAC export and no promise that the source sample rate is preserved. PCM can be larger than a compressed input. Actual output must fit within 24 MiB and the 120 second duration limit. For this 16.640 second example, a mono sample payload is 1597440 bytes before WAV container overhead. Check the actual measured duration, frames, channels, and bytes rather than relying on the estimate. The audio file is what a collaborator needs to hear or place in a video timeline.

Save settings JSON retains the editing choices, title, note, and platform context, not the sound, source filename, or measurements. Its versioned format uses kind smtasker-audio-trim, version 1, and fields for platform, title, start, end, fadeIn, fadeOut, gainDb, and note. A null end means the prepared source end. Title is a single line up to 80 codepoints; note allows up to 600 codepoints across 12 lines. Opening valid settings keeps the audio currently loaded, so verify the take and correct any range mismatch manually. Settings do not directly import into SMTasker and cannot restore an absent recording.

Keep the complete CSV, TXT, or copied review when you need the measured result as well as the choices. The styled PDF includes a waveform and metrics and attaches complete settings plus review JSON, but no audio or source filename. Its visible pages are rasterized; choose TXT or CSV for selectable review text. The authored title and note can appear in reports, so review those fields before sharing. Record the wording decision clearly, such as “keep Before and the complete final word,” without including private participant details. Save the original recording and your own source-to-revision mapping separately from these exports.

Check the instruction in the assembled video and keep a revision

Place the WAV with the route-marker footage in your video editor and review the whole sequence. A complete audio sentence can still refer to a sign before it appears or continue after the picture has moved somewhere else. Adjust the picture timing or audio placement in that editor; this tool does not inspect video, mix tracks, or change speed and pitch. Camera sound may also duplicate the voice or cover a quiet word. Check the exported video as well as the timeline preview. When the destination processes it, review that actual version rather than assuming a good local WAV guarantees the final presentation.

If the excerpt will be adapted for another context, use the Content Repurposing Planner to record the new brief and its required references. Do not reuse the same cut merely because its words are technically complete. A route direction may depend on a visible marker or a preceding explanation that another edit omits. Keep an external revision name linking the approved WAV to the intended video and its wording note. Return to the manual audio trimmer for another source range when the meaning or delivery changes, and repeat the listening check after each new preparation.

Before leaving, download the assets you need. There is no input upload or autosave, although normal site assets and the public example may load. Leaving releases decoded audio buffers and prepared output. Browser history may retain a local File in page memory for an explicit Restore audio action; reload requires reselecting it. That possibility is not a durable backup. For recurring supported account work, SMTasker is a separate workflow with schedules, limits, and logs, not a word-boundary detector or an importer for this settings file. Keep the recording, editorial approval, video assembly, and account routine distinct enough to review each honestly.

Questions about Keep Words Complete

Continue with another audio-editing guide

Apply the idea in the Audio Trimmer & Fade Editor, or continue with another focused audio guide. Browse more utilities in all free tools.