AUDIO TOOLS / VOICE STUDIO

Voice Delivery Analyzer

Explore the measurable patterns of speech—pitch movement, vocal energy, pauses, and rhythm—without guessing what a person actually feels.

◉ Private on-device analysis⌁ Pitch & energy timelines✓ No paid model required

Bring your voice

Upload a recording or capture a short microphone clip.

↥Drop an audio file hereor click to choose a recordingWAV, MP3, M4A, OGG, FLAC and other browser-supported formats
60 MB maximum · 8 minutes decoded duration · 256 MB decoded PCM safety limit

◉Record a quick sample

Use your microphone without uploading or saving a recording to any server.

Permission required · max 90 seconds

0:00 / 0:00

Start at (seconds)End at (seconds)Analysis channel

Analyze an excerpt or the full file. Auto uses the stereo sum unless phase cancellation substantially weakens it. Original playback is never changed.

Choose a file to begin.

Vocal delivery measurements

Direct measurements and estimates from the selected audio.

NO REPORT

Median detected pitch—Confident voiced frames · Hz

Pitch movement—10th–90th percentile range

Average energy—RMS dBFS · not LUFS

Internal pauses—Estimated silent intervals ≥0.28 s

Active sound—Amplitude-based, not speech ID

Rhythmic activity—Energy pulses/min · not words/min

Analysis channel: —Pitch reliability: —

Descriptive delivery notes

NOT EMOTION RECOGNITION

A successful analysis will summarize pitch movement, level variation and pauses here.

Optional AI delivery-writing assistant

Draft practical, non-diagnostic notes from locally measured voice features. No speech, transcript, filename or microphone data is sent to the text model.

Optional model is not loaded. You can analyze audio and export reports without it.

First-response diagnostics

No AI inference performed yet.

Inspect the recording

Select points on a timeline to jump to the audio.

Energy over timeWaiting for a recording

Relative level is influenced by recording gain, distance and compression—not just vocal expression.

Detected pitch contourWaiting for a recording

Fundamental frequency is estimated only on frames with sufficiently periodic, audible audio. Polyphonic sound and noise reduce accuracy.

Pauses to review

No analysis yet

Longer low-level gaps will appear here, with buttons to listen around each gap.

Optional speaking-rate estimate

Add an accurate transcript to estimate actual words per minute.

Transcript of the analyzed audio section

Counted words ÷ full excerpt duration, including pauses. Not automatic transcription. This may be unsuitable for languages without word boundaries.

— WPM

Export the analysis

Share measurements without sharing the source recording.

Reports include measured delivery features and cautionary context; transcript text itself is not included in exports.

✦

Delivery, not feelings.

Higher volume does not establish anger. A rising pitch does not establish happiness. These are acoustic traits, not a window into someone’s emotions or intentions.

SoundStays on device

ProcessingLocal JavaScript

AI modelNot required

ExportsJSON / CSV / PNG

When results need caution

Works best with one clean speaking voice. Background music, overlapping speakers, phone filters, whispered speech and distant microphones can distort estimates.

Pauses are level-based estimates; a quiet word may be mistaken for silence. Pitch tracking is unreliable for some vocal styles and recordings.

How it works

Web Audio decodes the file locally. A background worker measures amplitude, fundamental-frequency candidates and silence intervals. Optional WPM uses only the transcript you supply.

Despite the category name “AI Voice Emotion Analyzer”, this version uses transparent signal-processing methods; it does not claim a trained emotion classifier.

Frequently asked questions

Can this tool tell whether a person is angry, happy or afraid?

No. A recording can show vocal pitch, volume, rhythm and pausing, but these cannot reliably identify actual emotions. The tool reports the observable acoustics without judging the person.

Does speaking rate come from automatic speech recognition?

No. Without a matching transcript, only an envelope-based rhythmic-pulse estimate is displayed. Accurate words per minute requires actual words, which you can supply manually.

What happens with unusual stereo phase or channel differences?

Auto mode compares the first two channels and uses a mono sum unless destructive phase cancellation makes the result much quieter. You can override the analysis channel. This does not alter your uploaded file or playback.

Is my recording sent to an AI service?

The audio analysis runs locally with JavaScript. Optional AI delivery-writing guidance can download a text model from a CDN after your consent; only acoustic measurements—not audio, transcripts or filenames—are provided to the local model. No remote inference service is used.

Why is pitch missing?

Pitch requires an audible periodic signal. Whispering, background music, overlapping voices, distortion and weak audio can make pitch detection unreliable, so this tool leaves the metric blank rather than inventing one.

Using Voice Delivery Analyzer

Import or record speech, select an excerpt and channel mode, then inspect delivery measurements. Supply a matching transcript only if you want a rough words-per-minute estimate. Export the needed reports and interpret them in context.

Optional text guidance requires an explicit AI model download, but the acoustic analyzer remains fully usable without AI. The model never receives the raw audio or transcript. Despite the source package's emotion-analyzer name, this tool does not detect emotions, personality, deception or clinical state. Pitch periodicity is not a probability of correctness; rhythmic pulses are not syllable counts. It performs no ASR. Microphone access needs secure context/permission. File/duration/decoded-memory limits apply; transcript text is excluded from exported reports, while acoustic results can still be sensitive.

Review voice delivery measurements without inferring emotion

Provide a supported recording and, if useful, a transcript for the available comparison. The analyzer reports acoustic observations such as pitch, level, pauses and rhythm; it does not infer emotion, personality, deception, health or the correctness of spoken claims.

Example workflow

Choose a short recording and listen before relying on numeric measurements.

Review pitch, energy, pause and rhythm summaries in context; add transcript text only for the supported transcript-assisted words-per-minute view.

Export or save the report you need, then treat it as a description of this recording rather than a diagnosis or judgment about the speaker.

Authored example: Compare two readings of the same neutral sentence recorded at different distances from a microphone. Look at level and pause differences, but do not label them as confidence, stress or emotional state.

Limits and privacy: The tool has no emotion classifier and performs no automatic speech recognition. Microphone capture requires a secure context and permission. Transcript text is excluded from exported reports, while acoustic measurements may still be sensitive. File, duration and decoded-memory limits apply.

Voice Delivery Analyzer questions

Can the measurements tell how a speaker feels?

No. Pitch, level, pause and rhythm measures do not establish emotion, personality, deception, health or intent.

Does the analyzer transcribe speech automatically?

No. It performs no ASR. Transcript-assisted words-per-minute analysis requires text supplied through the supported input.