Six common risk groups can derail an online audio-effects job: the file may not decode, the output may exceed limits, samples may clip, loudness may be misunderstood, export metadata may disappear, and repeated processing may degrade the result.
These are different problems. A browser can reject encoded audio before an effect runs; a completed job can still sound wrong because the selected operation does not match the intended sound; and a valid WAV can differ from its source in codec and metadata. The Audio Effects Online workflow reads, decodes, processes, encodes, and creates downloads in the current browser tab. Audio is not uploaded to Lizely or stored in an account, and there is no server render or hidden AI transformation.
Separate failed jobs from misleading results
Within this workflow, a missing result can mean the input could not be decoded or the processed result exceeded a resource limit. A file that downloads but sounds wrong points toward the effect, amount, clipping, or a mismatched expectation. A file that downloads with missing metadata is still a successful export, but it is not a source-container-preserving conversion.
| Symptom | Likely cause | Useful response |
|---|---|---|
| No result appears | Browser decoding failed, or decoded or output data exceeded a limit | Confirm that the source is browser-decodable, then check the 50 MiB and channel-sample limits. |
| Echo or reverb exceeds the output budget | The complete added tail raises the output above 30 million channel samples | Use a shorter input or an effect without an added tail; the tool will not truncate the result. |
| Peaks sound flattened | Dry and delayed samples overlap and clip | Lower the amount and inspect the output peak before using the file. |
| Normalized files still differ in perceived loudness | Normalization targets sample peak rather than perceived loudness | Do not treat it as an equal-loudness or broadcast-compliance step. |
| Tags or artwork are missing | The export is a new PCM16 WAV | Keep the original file and its metadata separately. |
| A later chained pass sounds worse | Artifacts were exposed, or repeated clipping or normalization changed dynamics | Restart from the original and compare one deliberate change at a time. |
The format list is broader than the set of files every browser can open. MP3, WAV, M4A, AAC, Ogg, WebM, and FLAC names are accepted, yet actual codec and profile support varies by browser and operating system. A recognized filename does not guarantee that a particular encoded file will decode, and decoding failure creates no result. The tool uses the browser’s Web Audio implementation, whose decoding and AudioBuffer contract is specified in the W3C Web Audio API 1.1.
Check the input and output limits before processing
A validated job must satisfy all of these boundaries:
- One browser-decodable encoded file no larger than 50 MiB.
- One to eight equal-length channels and a sample rate from 8,000 to 192,000 Hz.
- A whole-number effect amount from 0% through 100%.
- No more than 30 million channel samples, counted across all channels, in either the decoded input or the complete effect output.
Input and output budgets are not interchangeable. Echo and reverb allocate their complete tails before processing, so a source can pass the input limit and still fail the output limit. When that happens, nothing is truncated and no file is created. Changing the file, effect, or amount also invalidates older work; stale jobs are prevented from replacing newer state, but the newly generated result still needs its own inspection. Local processing removes an upload step, not browser decoding or resource limits.
Know which effect can create a false expectation
Each run applies one deterministic operation. The amount control blends or scales that operation; it does not select an impulse response, preset, or mastering target.
| Effect | Exact operation | Potential mismatch |
|---|---|---|
| Echo | At 100%, every input sample is retained and a half-strength copy is added 200 milliseconds later; lower amounts scale the delayed copy linearly. | Dry and delayed samples can overlap and clip. It is one simple, repeatable delay, not a room or hardware emulation. |
| Lightweight reverb | Copies are added at 29.7, 37.1, and 41.1 milliseconds with weights of 0.50, 0.35, and 0.25 multiplied by the selected amount. | It is an early-reflection design, not a real room, plate, spring, convolution, or professional reverberation tail. |
| Reverse | Sample frames are read in the opposite order and blended with the original according to the amount. | Expect a blend rather than an independent dry layer; reversing a source with existing clipping does not repair it. |
| Fade in or fade out | Fade in scales from the selected dry proportion at the beginning to full level at the end; fade out performs the inverse as a linear ramp. | It provides a full-range linear fade rather than a custom envelope or multi-stage shape. |
| Normalization | The largest absolute decoded sample is scaled toward a 0.95 peak. | It controls peak level, not perceived loudness, and silent input remains silent. |
Clipping deserves special attention because a successful file does not guarantee an undistorted result. When dry and delayed samples overlap, finite output values are constrained to the legal -1 through +1 range before WAV quantization. That prevents illegal values, but it cannot reconstruct peaks that were flattened by overlap. The reverb coefficients are disclosed product choices, not measurements of a concert hall, studio, plate, spring, impulse response, acoustic material, or branded processor. Its result can add space or repetition for a quick sound-design draft, but it does not model room dimensions, frequency-dependent decay, stereo diffusion, damping, pre-delay, convolution, or a professional reverberation tail.
Normalization has a separate failure mode: confusing peak control with loudness matching. Scaling toward a 0.95 sample peak does not target LUFS, perceived volume, broadcast compliance, dynamics, inter-sample true peak, or equal loudness across recordings. A quiet recording and a dense recording can therefore receive very different perceptual changes.
Run an audio-effects job safely in the browser
- Protect the source. Keep an untouched copy of the original. The tool keeps the original separate from the processed export, but another copy protects your recovery options if you later chain several edits.
- Choose one input. Select one browser-decodable encoded file no larger than 50 MiB. If the browser cannot decode it, the tool produces no result; changing the extension does not repair an unsupported codec or profile.
- Select one operation. Choose echo, lightweight reverb, reverse, fade in, fade out, or normalization, then set a whole-number amount from 0% through 100%. One run applies one transformation.
- Apply it locally. Processing occurs in the current tab with no server render. For echo or reverb, allow for complete tail allocation; if the output exceeds 30 million channel samples, the job creates no file rather than truncating it.
- Inspect before downloading. Check the reported channel count, sample rate, input and output frame counts, output peak, and exact WAV byte size. Listen for clipped transients or an unintended blend. Treat an output peak at the legal boundary as a reason to check for flattened peaks.
- Download and preserve the result. Save the complete PCM16 WAV and keep the untouched original. To build a chain, download and inspect that WAV before using it as a later input. Repeated clipping or normalization can change dynamics, and repeated processing may reveal source artifacts.
The byte-size reading refers to the new interleaved, little-endian PCM16 RIFF/WAVE file, not the original file with its original container and compression. For 16-bit PCM, Microsoft’s WAVEFORMATEX structure gives mono two bytes per frame and stereo four bytes per frame. The tool reports the exact complete WAV byte size after processing.
Account for format and metadata changes
A completed download preserves the processed channel samples and sample rate, but it is still a new file rather than a source-container copy. Floating samples are quantized to signed 16-bit PCM. The export does not preserve the source codec, bitrate, compression quality, tags, artwork, chapters, cue points, loop markers, or container metadata.
If any of those details matter, retain the original and its metadata separately. If you need several operations, chain them deliberately: download and inspect one result, then use that WAV as the input for the next run. The later run may expose source artifacts, while repeated clipping or normalization can alter dynamics, so compare each stage with the untouched source.
Make one controlled change when a result disappoints
Begin again from the untouched original rather than trying to compensate for an uncertain earlier pass. If the job was rejected, correct the decoding or budget issue. If it completed but sounds wrong, change only one variable—the effect or amount—and compare the frame counts, peak, and new WAV before proceeding. This makes the cause clearer and prevents normalization, clipping, and unintended blending from being mistaken for one another.