Skip to content

Editorial Standard — breathing session (guided breath pattern + audio)

Purpose

A breathing session pairs a short intro paragraph with a visual breath-pacer that guides the user through a named pattern (box, 4-7-8, extended exhale, physiological sigh, diaphragmatic). The session is the fastest-acting nervous-system intervention in the app: it can interrupt the tinnitus-to-anxiety feedback loop within 30 seconds (physiological sigh) or 2-4 minutes (box/extended exhale). It is not a meditation session -- the pacer does the work. The intro copy and audio narration provide context, framing, and motivation to complete the rounds.

Breathing sessions serve two distinct use-case clusters: spike/acute relief (physiological sigh, crisis breathing variants) and general relaxation/sleep support (box, 4-7-8, diaphragmatic). The register and framing differ substantially between these clusters.

Voice register

Spike/acute variants (physiological sigh, box in crisis): acute register. Directive, specific, fast to the point. The user is already in distress and does not need a preamble. Name the technique, say why it works for spikes, give the instruction. "Breathe in twice through your nose. Hold briefly. Long exhale through your mouth. That interrupts the spike response." No warmth before orientation.

Relaxation variants (4-7-8, diaphragmatic, extended exhale, sleep): early habituation register. Companion-forward, still, warm. The intro frames the session as a practice with a concrete mechanism, not as a "quick stress-relief exercise." The voice creates a small amount of space before the pacer begins.

In both clusters: the knowledgeable guide explains what is happening neurologically, then the pacer takes over. Do not narrate emotional states ("you will feel calmer"). Describe physiological processes ("the extended exhale activates the parasympathetic branch").

Evidence / IP

Sourced from docs/session-audio-protocols-ip-guide.md. Key technique-level evidence:

  • Diaphragmatic/coherent breathing: public domain. Lehrer and Gevirtz coherent breathing at 5 bpm (PMC7578229). Frame as interrupting the tinnitus-to-anxiety feedback loop via extended exhale.
  • Box breathing (4-4-4-4): public domain/unprotectable. Eyes-open and commute-compatible. Suitable for sustained settling after initial crisis breathing.
  • Extended exhale (4 in, 8 out): most evidence-based pattern for acute calming (evidence via Lehrer/Gevirtz lineage). The ratio itself is not proprietary.
  • 4-7-8 breathing: technique unprotectable; Weil's text/brand protected. Write own script; attribute to pranayama roots. Clinical reference: PMC9277512.
  • Physiological sigh: Yadin, Feldman, Huberman 2023 (Cell Reports Medicine, PMC9873947). Double inhale through the nose, long exhale through the mouth. Fastest acute sympathetic downregulation available without equipment.
  • 4-7-8 breathing hold phase: the hold_full phase uses 7 seconds per the technique. The pattern array in the schema uses seconds: 0 to skip a phase (e.g., no hold in diaphragmatic variants) -- use zero, not omission.

IP rule: all techniques above are public-domain or unprotectable. Write own narration. Do not reproduce Weil's scripted text for 4-7-8. For VA-derived scripts (physiological sigh context), material is US government public domain.

Length / reading level

  • Intro paragraph: 2-4 sentences. States the technique by name, its mechanism in plain language, and what the pacer will guide the user through. No wellness abstractions.
  • Kicker (optional overline): 1-2 words maximum. "RELAXATION", "CRISIS RESPONSE", "SLEEP PREP". All caps, spare.
  • Audio intro asset: voiced reading of the intro paragraph plus brief technique orientation. Target 30-60 seconds before the pacer starts.
  • Audio outro asset: 2-3 sentences. Acknowledges the user finished the rounds; names what just happened physiologically. Does not close with "well done" or celebratory language -- closes with the mechanism ("your nervous system has shifted into its recovery state").
  • Reading level: Grade 8 or below. The user may be mid-spike reading this.

Editorial-required elements

  1. Pattern fidelity. The pattern array must accurately encode the named technique. A 4-7-8 session with the wrong second counts for each phase is a clinical error, not a copy error. Verify the phases against the IP guide before publishing.

  2. Mechanism in the intro, always. The intro paragraph must state why this breathing pattern works, not just that it does. "The extended exhale activates the parasympathetic response and lowers cortisol" is mechanism. "This breathing exercise is calming" is not.

  3. Tinnitus-specific framing where applicable. The IP guide explicitly calls out the frame: "an extended exhale interrupts the tinnitus-to-anxiety feedback loop." Use that frame in relevant variants -- it is clinically accurate and directly relevant to the user's situation. Diaphragmatic breathing variants without the tinnitus frame are a missed opportunity.

  4. Spike variants: no preamble. The intro for physiological sigh and crisis breathing variants must be usable mid-spike. Maximum 3 short sentences. If the user needs to read 6 sentences before they can start, the session has failed for its primary use case.

  5. Outro: mechanism not reassurance. The outro names the physiological process that just occurred, not how the user should feel. "Your heart rate variability has increased" is mechanism. "You should feel calmer now" is reassurance and also an implicit pressure if the user does not feel calmer.

  6. Obeys ai-patterns-en.md. Intro and outro copy must obey all of ai-patterns-en.md -- no em dashes in produced copy, no wellness clichés, no breathiness, no toxic positivity.

  7. Audio scripts live under narration: {intro, outro}. The render (audio_pipeline) reads the audio intro/outro from a body narration block with intro: and outro: segment lists ({text, pause_after_ms}), plus register: breathing. The loop's spoken phase-cues ("Breathe in / Hold / Breathe out", with a soft count on phases ≥ 5s) are generated from content.pattern -- never authored. The on-screen content.intro is independent of the audio intro script.

Intro requirements (narration.intro)

The spoken narration.intro (NOT the on-screen content.intro) must, in one natural flow:

  1. Explain the pattern using the actual content.pattern seconds — e.g. "Breathe in for four, hold for seven, breathe out for eight."
  2. Tell the user they can tap the round number to change how many rounds they do.
  3. Tell the user they can tap the sound icon to add a background sound that masks their tinnitus.
  4. End by saying the exercise is about to begin, and set pause_after_ms: 3000 on the last intro segment — the 3-second prep pause before the loop. (validate.py enforces this pause; the other three elements are scored by the review gate.)

Two intros, two limits: the on-screen content.intro (the paragraph above the CTA) is capped at ≤120 chars (it names the technique, nothing more). The spoken narration.intro is the longer guided flow above and is NOT length-capped.

Examples

Good -- intro for box breathing (crisis variant):

Box breathing slows the sympathetic response by spacing inhale, hold, exhale, and hold across equal four-second counts. Follow the pacer for six rounds. The pattern was developed for high-stress training contexts -- it works by giving the nervous system a predictable rhythm to anchor to.

Why this works: names the technique, states the mechanism (sympathetic response, predictable rhythm), references a real grounding (high-stress training contexts from the military/first-responder tradition), and tells the user exactly what to do (follow for six rounds). No wellness-abstract language.

Bad -- intro for a relaxation breathing session:

❌ Take a moment to connect with your breath and find your inner calm. This gentle breathing exercise will help you manage your tinnitus and feel more at peace as you navigate this challenging journey.

Why it fails: "find your inner calm", "manage your tinnitus", "feel more at peace", and "navigate this challenging journey" are all banned under ai-patterns-en.md (Naluma-voice additions). The intro contains no mechanism. It could have been written for any wellness app without knowing anything about tinnitus or breathing physiology.

Pre-publish audio QA

Before a locale's audio goes published, do a sampled listen-check on each rendered <locale>.{intro,loop,outro}.m4a:

  • Right voice for the locale (see audio_pipeline/.../voices.toml).
  • Right language — spot-check the spoken intro/outro and the loop phase-cues are in the target language.
  • No artifacts — no clipping, abrupt cuts, or wrong-pace pacer cues; the loop counts down (3, 2, 1) and the outro has its 2 s lead silence.

The breathing render emits no transcript (the loop's spoken cues are generated from content.pattern), so this human listen-check is the QA backstop — there is no transcript-vs-script assertion for breathing (no ASR by design).