Recording & Self-Monitoring¶

You cannot train a voice you cannot hear. Real-time self-perception is unreliable — bone conduction makes your own voice sound deeper to you than it does to others, and stress alters perception further. The fastest way out of this gap is systematic recording and listening back, with apps that show pitch and formants in real time as a secondary aid.

This page covers how to record yourself usefully, what to listen for, the major app inventory, and how to keep a longitudinal log of progress.

Why your voice sounds different on playback¶

The single most important fact for a beginner is that the voice you hear in your head is not the voice anyone else hears, and the gap is not small. The recording is the listener-accurate signal. The internal experience is the illusion.

Bone conduction, briefly¶

When you speak, sound reaches your cochlea by two paths: through the air (the same path every other listener uses) and through the bones of your skull. The bone-conducted path preferentially transmits low-frequency energy, so your self-perception is biased toward the low end — fundamentals and lower harmonics are emphasized. The air-only signal a microphone captures has none of that extra bass.

This is why first playback "sounds higher" or "more feminine" than expected. The recording is not flattering you; it is removing the bone-conducted bass that has been coloring your self-perception since you started speaking.

Cognitive load distorts perception further¶

Under load — formulating a sentence, holding eye contact, navigating a conversation — auditory self-monitoring degrades systematically. Trainees under-perceive their own pitch drops, resonance collapses, and creak. A voice that "felt fine" in conversation can sound noticeably masculine on review.

Practical implication¶

Real-time perceptual self-correction has a ceiling. Past that ceiling, progress requires recording and listening back with the same ears the listener is using. Everything else on this page follows from that.

Recording technique¶

A modern phone mic is more than adequate for self-monitoring. The mistakes are about geometry, not gear.

Phone placement¶

Distance: roughly 6–12 inches (15–30 cm) from the mouth. Closer produces proximity-effect bass boost and risks plosive distortion (popped Ps and Bs); farther brings in room reverb and lowers signal-to-noise.
Angle: off-axis by ~30–45°, not pointed straight at the lips. Off-axis placement avoids direct plosive blast and is more forgiving of breath noise.
Stability: put the phone on a surface or stand. Holding it changes distance every breath and contaminates the recording with handling noise.
Mic location: know where the bottom mic on your specific phone is — several modern phones have multiple mics and the "primary" one is not always where you'd guess. Cover the wrong one with a finger and you've cut your usable signal.

Room treatment basics¶

You are not building a studio. You are avoiding the worst rooms.

Soft furnishings dampen reverb. A bed, curtains, a rug, a couch, and a bookshelf full of clothes or books will all kill enough reflections to make a recording usable.
Hard rectangular rooms are the enemy. Bathrooms, kitchens, and empty bedrooms with parallel walls produce flutter echo and standing waves that color the spectrum unevenly. That coloration corrupts both perceptual judgments and any spectrograph-based analysis.
A workable cheap fix: record sitting on the bed facing into a half-open closet full of clothes. The clothes act as a broadband absorber.
Avoid recording outdoors for analytic comparison — wind and traffic add non-stationary noise that makes month-to-month comparison unreliable.

Headphones for playback, not while recording¶

Playback through headphones is essential. Phone speakers roll off the low end and color the mids unpredictably; what sounds "fine" on the built-in speaker may be muddy or thin in real listening conditions. Use closed-back headphones or decent earbuds for playback.
Do not wear headphones while recording yourself speaking. Monitoring your own voice through headphones loops a delayed, colored version back into your ears alongside the bone-conducted path. This destroys the feedback you're trying to calibrate. Record raw; listen back afterwards.
The narrow exception is professional sidetone monitoring with very low latency; this is not what a phone provides.

File naming and dates¶

A naming convention is the single intervention that makes longitudinal comparison possible. Files named Recording 47.m4a are useless six months in.

Default pattern: YYYY-MM-DD_short-label.m4a — e.g. 2026-04-29_rainbow-passage.m4a, 2026-04-29_casual-call-sample.m4a.
ISO date prefix sorts correctly in any file browser.
Keep labels short and consistent: rainbow, casual, read-aloud, vowel-sustained, pitch-glide.
Optional suffix for session context: _warmed, _cold, _post-coffee, _tired. Useful when comparing best-case versus realistic samples.
Store in a single folder synced to a cloud backup. Do not trust a phone's voice-memo app alone with a year of data.

What to listen for¶

Listening is the skill the apps cannot replace. Train your ear on these specific failure modes.

Pitch consistency across a sentence¶

Does pitch hold through the sentence, or does the last syllable drop into vocal fry (creak, glottalization)? Sentence-final fry is a near-universal pattern in English, and in trans voice work it's a high-yield thing to catch because the listener's last impression is weighted heavily.
Listen for unintended downward pitch drift across longer utterances — pitch slowly descending as breath support fades. See Breath Support.

Resonance shifts under cognitive load¶

Set up a recording where you read or recite something easy (the warm-up voice) and then transition into something harder (explaining a concept, recounting an event, problem-solving aloud).
Listen specifically for the moment the voice collapses back to chest resonance mid-sentence as cognitive attention shifts to content. The trained voice is the first thing to go under load; the recording shows exactly where.

Sentence-final patterns¶

Does intonation hold or drop at the end of statements? A heavy sentence-final drop into fry reads as masculine even at a feminine fundamental.
Does pitch rise on questions and lists? Inconsistent intonation reads as flat or "off" even when fundamental frequency is in range. See Intonation & Prosody.

The "every five minutes" rule¶

A specific protocol for catching the moment trained voice slips:

During a practice session, set a recurring 5-minute timer.
At each timer, record ~30 seconds of casual speech — not a prepared passage. Talk about your day, narrate what you're doing, anything unscripted.
Review the timeline of clips afterward. The slip is almost always visible: the first few clips hold; somewhere in the session resonance collapses or pitch drops; from that point on the voice is no longer trained-voice.
The point of identifying the slip is to learn its triggers (fatigue, task complexity, emotional content) and to set practical session duration based on actual sustainable performance, not aspiration.

Apps and tools¶

Each tool has a specific job. Trying to do everything in one app is a recipe for plateauing. Treat the apps as feedback, not as graders.

Comparison¶

App	Platform	Cost	What it's good for
Pitch Reader	iOS / Android	Paid (low)	Pitch-matching against target notes; F0 drills and sustained-target practice
Friture	Linux / Windows / macOS	Free, OSS	Real-time spectrograph; watching F1/F2 shift during resonance work
Christella VoiceUp	iOS / Android	Paid (mid)	Structured trans-voice curriculum (pitch glides, resonance drills, prosody)
Voice Tools	iOS / Android	Free / freemium	Quick F0 check and basic spectrogram during the day, no curriculum
Praat	Windows / macOS / Linux	Free	Research-grade acoustic analysis, scriptable, longitudinal comparison

Pitch Reader¶

Pitch-matching against target notes. The training mode shows a moving pitch line against a held target and scores how long you stay in the target band.

Best use: F0 drills, glide exercises, hitting and sustaining a target fundamental. Pairs well with the pitch-matching drill in Exercises.
Limitation: no formant feedback. Treats pitch as the only variable, which is exactly the trap to avoid.

Friture¶

Real-time spectrograph and harmonic analyzer. Runs on a laptop with the built-in mic.

Shows formants F1, F2, F3 as horizontal bands along the spectrogram and the fundamental as the lowest narrow band, with harmonics stacked at integer multiples above it.
The serious option for resonance work: you can watch F1 and F2 shift in real time as you change tongue position and pharyngeal shape. See Resonance Fundamentals for what to aim for.
Free and open source; the trade-off is that it expects you to know what you're looking at. Pair with the spectrograph reading basics below.

Christella VoiceUp¶

Commercial voice-feminization training app with structured exercises (pitch glides, resonance drills, articulation, intonation patterns).

Strength: scaffolded curriculum for trainees who want a path rather than a toolbox.
Limitation: still an app. Not a substitute for a live SLP if budget allows one — see Working with an SLP.

Voice Tools¶

F0 measurement plus a basic spectrogram view. Sample-and-display rather than a full curriculum. Useful as a quick check during the day — "what's my fundamental right now?" — without launching a desktop tool.

Praat¶

Research-grade acoustic analysis (Boersma & Weenink, University of Amsterdam). The standard tool in SLP and phonetics literature for the last 30 years.

Pitch tracking, formant tracking, jitter, shimmer, harmonics-to-noise ratio, spectral tilt, intensity contours, scriptable batch analysis.
Steep learning curve and an aggressively unfashionable UI. Worth the effort for anyone doing serious longitudinal analysis or preparing samples for an SLP.

Spectrograph reading basics¶

A short orientation so the visualizations are not just pretty colors.

X-axis: time. Y-axis: frequency in Hz. Color or intensity: energy at that time-frequency bin (brighter = more energy).
Fundamental (F0): the lowest narrow horizontal band, plus a stack of equally spaced narrow bands above it (harmonics at 2×F0, 3×F0, etc.). The whole stack moves up and down together when pitch changes — that's the signature of F0 motion versus formant motion.
Formants (F1, F2, F3): broader horizontal bands of concentrated energy that sit on top of the harmonic stack. They reflect vocal tract shape, not vocal-fold vibration. Roughly: F1 tracks tongue height and pharyngeal opening; F2 tracks tongue front-back position; F3 tracks lip rounding and finer vocal-tract detail.
For trans voice work: F0 typically targets ~165–220 Hz, though the gendered-perception range is wider and overlaps with cis male range above ~150 Hz. Feminine resonance work raises F1, F2, and F3 by shortening the effective vocal tract (smaller pharynx, raised larynx posture, forward tongue). The spectrograph shows whether the change actually happened.

Limitations of app-driven training¶

Apps measure what they measure. F0 and formants are necessary but not sufficient for perceived gender. Prosody, voice quality (breathy/pressed/clear), articulatory precision, and speech rate all contribute and are not well-instrumented in consumer apps.
A voice can hit every F0 and formant target on the spectrograph and still read as masculine because of intonation pattern, sentence rhythm, or vocal weight.
Treat the apps as feedback, not as graders. The recording-and-listening loop is the primary signal; the apps quantify pieces of it.

Keeping a longitudinal log¶

The hardest part of voice training is that progress is invisible on a day-to-day scale and obvious on a month-to-month scale. The log is what makes the slow change legible.

Weekly recordings of a fixed passage¶

Pick one canonical passage and record it once a week — same passage, same conditions if possible.
The Rainbow Passage is the SLP-standard reading sample: roughly 330 words, phonetically balanced, written for voice and articulation assessment. It is public domain and used in nearly every research paper in the area, which makes self-comparison and comparison to literature norms easier.
Supplement with a sustained vowel (/a/ for ~5 seconds) and a short unscripted casual sample. The three-sample set covers steady-state phonation, prepared reading voice, and real-world speaking voice.

Comparing month-to-month¶

Do not compare week N to week N-1. The differences are mostly noise: time of day, hydration, sleep, hormones, ambient temperature, mic placement variance.
Compare week N to week N-4 or N-8. The differences at this scale are signal: F0 baseline shifts, formant target stability, prosody patterns consolidating.
Once a quarter, listen to a clip from three months ago back-to-back with this week's clip. This is where motivation comes from when daily practice feels stagnant.

Avoiding the "I sound terrible" loop¶

The first time a trainee hears her recorded voice she will almost always rate it more masculine and worse than her in-the-head version. This is not failure; this is the bone-conduction gap closing. The recording is listener-accurate; your internal experience was not.
Distinguish two things: (1) "this sounds worse than I expected" — almost universal early, not predictive of outcome; (2) "this sounds worse than last month" — actually informative, and usually wrong on inspection (compare clips, not memory).
A useful habit: when a recording feels devastating, save it anyway and listen again in a week. The reaction usually moderates. If it doesn't, see Troubleshooting & Plateaus.

Compare clips, not memories

Memory of how a recording sounded last month is dominated by the emotional response to it, not the acoustic content. The clip is still there. Listen to both back-to-back before concluding regression.

Recording for an SLP¶

If you are bringing samples to a first SLP appointment, three samples cover the standard intake. See Working with an SLP for what the appointment itself looks like and what to ask.

Sustained vowel — /a/ held steady for ~5 seconds at comfortable pitch and loudness. Used for jitter, shimmer, and harmonics-to-noise measures.
Reading passage — Rainbow Passage or equivalent, recorded straight through, normal reading voice (not "performed" feminine voice unless asked).
Casual conversational sample — 1–2 minutes of unscripted speech. Talk about your week, describe a recent activity. This is the sample the SLP will use to assess your real-world voice; the read passage alone is not representative.

Record all three in the same session, same room, same mic placement, named with the date. Bring the files on a phone or USB drive; do not rely on streaming from cloud during a session.

Bring the unflattering sample

The temptation is to bring only your best take. The SLP needs the realistic one. Include the casual sample even if you don't like how it sounds — that is the sample that actually informs the treatment plan.

Sources¶

Davies, S., Papp, V. G., & Antoni, C. (2015). Voice and communication change for gender nonconforming individuals: giving voice to the person inside. International Journal of Transgenderism 16(3): 117–159.
Boersma, P., & Weenink, D. Praat: Doing phonetics by computer. University of Amsterdam. https://www.fon.hum.uva.nl/praat/
Friture — real-time audio analyzer. https://friture.org/
Fairbanks, G. (1960). Voice and Articulation Drillbook. Origin of the Rainbow Passage as a standard reading sample.
Pörschmann, C. Influences of bone conduction and air conduction on the sound of one's own voice. Acta Acustica united with Acustica, 2000.
Hancock, A. B., Krissinger, J., & Owen, K. Voice perception and quality of life of transgender people. Journal of Voice, 2011.
Hancock, A. B. Recording-based and perceptual-rating intervention work in transgender voice. George Washington University, Department of Speech, Language, and Hearing Sciences (multiple papers).