@seams/voice-capture 0.67.1 → 0.68.0

package/dist/browser/chunks/{microphone-CU5gQG6w.js → microphone-BQVF_tVV.js}

@@ -1 +1 @@
-const e={assetBase:null,microphonePreference:null,styleNonce:null,label:`Voice capture`,place:`this page`,site:`this site`,recoveryHint:`Reload the page and try again.`,cacheName:`seams-voice-capture-models-v1`};let t=e;function n(n){let r=new URL(n.assetBase,globalThis.document?.baseURI);r.pathname.endsWith(`/`)||(r.pathname+=`/`),t={assetBase:r,microphonePreference:n.microphonePreference??null,styleNonce:n.styleNonce??null,label:n.label??e.label,place:n.place??e.place,site:n.site??e.site,recoveryHint:n.recoveryHint??e.recoveryHint,cacheName:n.cacheName??e.cacheName}}function r(){return t}function i(){let e=t.styleNonce;return(typeof e==`function`?e():e)??void 0}function a(e){if(!t.assetBase)throw Error(`Call configureVoiceCapture({ assetBase }) before you load a voice asset.`);return new URL(e,t.assetBase)}const o=3e4;var s=class extends Error{constructor(e,t,n){super(t,n),this.kind=e,this.name=`VoiceAssetLoadError`}};async function c(){try{return await caches.open(r().cacheName)}catch{return null}}var l=class{pending=null;progress={kind:`checking_cache`};listeners=new Set;constructor(e){this.asset=e}get size(){return this.asset.bytes}async download(e){e&&(this.listeners.add(e),e(this.progress));try{this.pending||=this.bytes().catch(this.failed);let e=await this.pending;return this.publish({kind:`ready`}),e}finally{e&&this.listeners.delete(e)}}failed=e=>{throw this.pending=null,this.progress={kind:`checking_cache`},e};publish(e){this.progress=e;for(let t of this.listeners)t(e)}async valid(e){if(e.byteLength!==this.asset.bytes)return!1;let t=new Uint8Array(await crypto.subtle.digest(`SHA-256`,e)),n=``;for(let e of t)n+=e.toString(16).padStart(2,`0`);return n===this.asset.sha256}async bytes(){let{fileName:e,sha256:t}=this.asset;this.publish({kind:`checking_cache`});let n=await c(),r=new URL(`${e}?sha256=${t}`,a(``)).href;try{let e=await n?.match(r);if(e){let t=await e.arrayBuffer();if(await this.valid(t))return t;await n?.delete(r)}}catch{}let i;try{i=await this.fetchBytes()}catch(e){if(!(e instanceof s)||e.kind!==`connection`)throw e;i=await this.fetchBytes()}if(n)try{await n.put(r,new Response(i,{headers:{"Content-Type":`application/octet-stream`}}))}catch{}return i}async fetchBytes(){let{fileName:e}=this.asset,t=typeof this.asset.label==`function`?this.asset.label():this.asset.label,n=new AbortController,i=setTimeout(u.bind(null,n),o);this.publish({kind:`downloading`,received:0});try{let c=await fetch(a(e),{credentials:`omit`,cache:`reload`,signal:n.signal});if(!c.ok||!c.body){let e=[408,429,500,502,503,504].includes(c.status);throw new s(e?`connection`:`unavailable`,e?`${t} download was interrupted. Check your connection and retry.`:`${t} is unavailable from ${r().site}. ${r().recoveryHint}`)}let l=c.body.getReader(),f=new Uint8Array(this.asset.bytes),p=0;try{for(;;){let e=await l.read();if(e.done)break;if(clearTimeout(i),i=setTimeout(u.bind(null,n),o),p+e.value.length>f.length)throw new s(`integrity`,`${t} size is invalid. Retry loading.`);f.set(e.value,p),p+=e.value.length,this.publish({kind:`downloading`,received:p})}}finally{await l.cancel().catch(d),l.releaseLock()}if(p!==f.length||!await this.valid(f.buffer))throw new s(`integrity`,`${t} integrity check failed. Retry loading.`);return f.buffer}catch(e){throw e instanceof s?e:new s(`connection`,`${t} download was interrupted. Check your connection and retry.`,{cause:e})}finally{clearTimeout(i),n.abort()}}};function u(e){e.abort()}function d(){}const f=new l({fileName:`ort-wasm-simd-threaded.wasm`,sha256:`3398c10d07d229bd91b364548e130e0e51a8e5704b88c7c083ebbeb78842dee2`,bytes:14239897,label:()=>`${r().label} engine`});var p=class{asset;session=null;constructor(e){this.asset=new l(e)}prepare(){return this.session||=this.load().catch(this.failed),this.session}async run(e){let t=await this.prepare(),{Tensor:n}=await import(`./ort.wasm.bundle.min-CWqlff_l.js`),r=new n(`float32`,e,[1,e.length]);try{return await t.run({samples:r})}finally{r.dispose()}}failed=e=>{throw this.session=null,e};async load(){try{let e=await import(`./ort.wasm.bundle.min-CWqlff_l.js`);e.env.wasm.numThreads=1,e.env.wasm.wasmPaths=a(``).href,e.env.wasm.wasmBinary=await f.download();let t=await this.asset.download();return await e.InferenceSession.create(t,{executionProviders:[`wasm`]})}catch(e){if(e instanceof s)throw e;let{label:t,recoveryHint:n}=r();throw console.error(`[${t}] inference engine could not start`,e),new s(`initialization`,`${t} could not start. Retry loading. If it still fails, ${n.charAt(0).toLowerCase()}${n.slice(1)}`,{cause:e})}}};const m=new p({fileName:`silero-vad.onnx`,sha256:`2623a2953f6ff3d2c1e61740c6cdb7168133479b267dfef114a4a3cc5bdd788f`,bytes:2327524,label:`Speech detection`});async function h(){return new _(await m.prepare(),await import(`./ort.wasm.bundle.min-CWqlff_l.js`))}const g={probability:async()=>1,clear:()=>{}};var _=class{state=new Float32Array(256);context=new Float32Array(64);constructor(e,t){this.session=e,this.ort=t}async probability(e){if(e.length!==512)throw Error(`Speech detection requires a 512-sample frame.`);let t=new Float32Array(576);t.set(this.context),t.set(e,64);let n=new this.ort.Tensor(`float32`,t,[1,576]),r=new this.ort.Tensor(`float32`,this.state,[2,1,128]),i=new this.ort.Tensor(`int64`,BigInt64Array.of(16000n),[]);try{let t=await this.session.run({input:n,state:r,sr:i});try{let n=t.output?.data,r=t.stateN?.data;if(!(n instanceof Float32Array)||n.length!==1||!Number.isFinite(n[0])||n[0]<0||n[0]>1||!(r instanceof Float32Array)||r.length!==256)throw Error(`Speech detection returned an invalid result.`);return this.state.set(r),this.context.set(e.subarray(448)),n[0]}finally{for(let e of Object.values(t))e.dispose()}}finally{n.dispose(),r.dispose(),i.dispose(),t.fill(0)}}clear(){this.state.fill(0),this.context.fill(0)}};function v(e,t){return t!==null&&e>=t?800:1500}function y(e){return e.trim().replace(/[.!?,;:]+$/,``).split(/\s+/).filter(Boolean).reduce((e,t)=>e+b(t),0)*200}function b(e){let t=e.match(/[A-Za-z]+/g)?.length??0,n=e.match(/[0-9]/g)?.length??0,r=e.match(/[.\-_@]/g)?.length??0;return Math.max(1,t+n+r)}const x=16e3,S=8*x;function C(e){let t=Math.min(e.length,S),n=new Uint8Array(44+t*2),r=new DataView(n.buffer);w(n,0,`RIFF`),r.setUint32(4,36+t*2,!0),w(n,8,`WAVEfmt `),r.setUint32(16,16,!0),r.setUint16(20,1,!0),r.setUint16(22,1,!0),r.setUint32(24,x,!0),r.setUint32(28,x*2,!0),r.setUint16(32,2,!0),r.setUint16(34,16,!0),w(n,36,`data`),r.setUint32(40,t*2,!0);for(let n=0;n<t;n+=1){let t=Math.max(-1,Math.min(1,e[n]??0));r.setInt16(44+n*2,Math.round(t<0?t*32768:t*32767),!0)}return n}function w(e,t,n){for(let r=0;r<n.length;r+=1)e[t+r]=n.charCodeAt(r)}let T={kind:`unread`};function E(e){return typeof e==`string`&&e.length>0&&e!==`default`&&e!==`communications`}async function D(){if(T.kind!==`unread`)return T;let e={kind:`system_default`};try{let t=await r().microphonePreference?.read();if(typeof t==`object`&&t&&`kind`in t&&t.kind===`device`){let n=`deviceId`in t?t.deviceId:null;E(n)&&(e={kind:`device`,deviceId:n})}}catch{}return T.kind===`unread`&&(T=e),T}function O(e){let t={channelCount:1,echoCancellation:!0,noiseSuppression:!0,autoGainControl:!0};return e.kind===`device`&&(t.deviceId={exact:e.deviceId}),{audio:t,video:!1}}async function k(e){if(E(e.deviceId))return{kind:`device`,deviceId:e.deviceId};if(e.groupId)try{let t=await navigator.mediaDevices.enumerateDevices();for(let n of t)if(n.kind===`audioinput`&&n.groupId===e.groupId&&E(n.deviceId))return{kind:`device`,deviceId:n.deviceId}}catch{}return{kind:`system_default`}}async function A(e){e.throwIfAborted();let t=await D();e.throwIfAborted();let n,r=t.kind===`device`?`preferred`:`default`;try{n=await navigator.mediaDevices.getUserMedia(O(t))}catch(i){if(e.throwIfAborted(),t.kind!==`device`||!(i instanceof DOMException)||i.name!==`OverconstrainedError`&&i.name!==`NotFoundError`)throw i;n=await navigator.mediaDevices.getUserMedia(O({kind:`system_default`})),r=`replacement`}let i=j.bind(null,n);e.addEventListener(`abort`,i,{once:!0});try{e.throwIfAborted();let t=n.getAudioTracks()[0];if(!t||t.readyState!==`live`)throw new F;let i=t.getSettings(),a=await k(i);if(e.throwIfAborted(),t.readyState!==`live`)throw new F;return{stream:n,preference:a,selection:r,label:t.label||`Default microphone`,processing:{noiseSuppression:i.noiseSuppression??null,echoCancellation:i.echoCancellation??null,autoGainControl:i.autoGainControl??null,sampleRate:i.sampleRate??null}}}catch(e){throw i(),e}finally{e.removeEventListener(`abort`,i)}}function j(e){for(let t of e.getTracks())t.stop()}async function M(e){e.preference.kind!==`device`||e.selection===`replacement`||await P(e.preference)}async function N(){await P({kind:`system_default`})}async function P(e){T=e;try{await r().microphonePreference?.save(e)}catch{}}var F=class extends Error{constructor(){super(`The microphone disconnected or stopped. Check your microphone and try again.`),this.name=`VoiceMicrophoneInterruptedError`}};function I(e){return e.noise===`high`?`Background noise is high. Move closer to the microphone and try again.`:`Audio too brief or quiet: ${(e.voicedMs/1e3).toFixed(2)} of ${(e.requiredVoicedMs/1e3).toFixed(2)} seconds detected.`}var L=class extends Error{constructor(e){super(I(e)),this.quality=e,this.name=`InsufficientVoiceAudioError`}};const R=.006,z=8e3;function B(){if(V(),!navigator.userActivation.isActive){let{label:e,place:t}=r();throw Error(`Open ${t} and select Record to use ${e}.`)}}function V(){if(document.visibilityState!==`visible`||!document.hasFocus()){let{label:e,place:t}=r();throw Error(`Keep ${t} open and focused to use ${e}.`)}}function H(){return Error(`${r().label} recording was cancelled.`)}function U(e){return e.detector===`level`?async()=>g:h}async function W(e,t,n={}){let r=n.expectedSpeechMs??null,i=U(n);if(e.aborted)throw H();let a=crypto.randomUUID();t({kind:`opening`});let o=await A(e),{stream:s}=o;try{if(e.aborted)throw H();let n=await new G(s,e,t,r,i).result.finally(t.bind(null,{kind:`processing`}));if(e.aborted)throw H();let c=await new OfflineAudioContext(1,x,x).decodeAudioData(n.bytes),l=new OfflineAudioContext(1,Math.ceil(c.duration*x),x),u=l.createBufferSource();u.buffer=c,u.connect(l.destination),u.start();try{let r=await l.startRendering();try{if(e.aborted)throw H();let s=r.getChannelData(0),c=await K(s,e,await i()),l={attemptId:a,capturedAtMs:Date.now(),durationMs:r.duration*1e3,recordingMs:n.recordingMs,endedBy:n.endedBy,voicedMs:c.voicedMs,peakFrameRms:c.peakFrameRms,levelThreshold:R,requiredVoicedMs:600,microphone:o.label,microphoneSelection:o.selection,processing:o.processing,backgroundRms:c.backgroundRms,speechToBackgroundDb:c.speechToBackgroundDb,noise:c.noise};if(t({kind:`measured`,quality:l}),c.kind===`insufficient_audio`)return{kind:`insufficient_audio`,quality:l};let u=s.slice(Math.max(0,c.first-3200),Math.min(s.length,c.last+3200));return await M(o),e.aborted&&(u.fill(0),e.throwIfAborted()),{kind:`captured`,samples:u,quality:l,lastSoundAtMs:n.lastSoundAtMs}}finally{q(r)}}finally{q(c)}}finally{for(let e of s.getTracks())e.stop()}}var G=class{recorder;chunks=[];timer=null;completion=Y();startedAtMs=performance.now();ended=null;activity={kind:`waiting`,activeMs:0};audioContext=null;worklet=null;detector=null;pending=Promise.resolve();queuedFrames=0;closed=!1;result=this.completion.promise;constructor(e,t,n,r,i){this.stream=e,this.signal=t,this.onCapture=n,this.expectedSpeechMs=r,this.createDetector=i,this.recorder=new MediaRecorder(e),this.recorder.addEventListener(`dataavailable`,this.data),this.recorder.addEventListener(`stop`,this.stopped),this.recorder.addEventListener(`error`,this.failed);for(let t of e.getAudioTracks())t.addEventListener(`ended`,this.interrupted),t.addEventListener(`mute`,this.interrupted);t.addEventListener(`abort`,this.cancelled,{once:!0}),this.timer=setTimeout(this.startupFailed,z),this.start()}async start(){try{if(await this.startPauseDetection(),this.signal.throwIfAborted(),this.startedAtMs=performance.now(),this.closed)return;this.recorder.start(),this.timer!==null&&clearTimeout(this.timer),this.timer=setTimeout(this.stop.bind(this,`limit`),z),this.onCapture({kind:`listening`})}catch(e){this.cleanup(),this.completion.reject(e instanceof Error?e:Error(`Speech detection could not start.`))}}data=e=>{this.chunks.push(e.data)};stop=e=>{this.recorder.state!==`inactive`&&(this.ended={endedBy:e,recordingMs:performance.now()-this.startedAtMs,lastSoundAtMs:this.activity.kind===`speaking`?this.activity.lastSoundAtMs:null},this.recorder.stop(),this.stopTracks())};cancelled=()=>{if(this.recorder.state===`inactive`){this.cleanup(),this.completion.reject(H());return}this.stop(`stream_ended`)};startupFailed=()=>{this.cleanup(),this.completion.reject(Error(`Speech detection startup timed out. Please try again.`))};detectorFailed=e=>{this.cleanup(),this.completion.reject(e instanceof Error?e:Error(`Speech detection failed.`))};failed=()=>{this.cleanup(),this.completion.reject(Error(`${r().label} could not record this microphone.`))};interrupted=()=>{this.cleanup(),this.completion.reject(new F)};stopped=()=>{let e=this.ended;if(this.cleanup(),this.signal.aborted){this.completion.reject(H());return}if(e===null){this.completion.reject(new F);return}new Blob(this.chunks,{type:this.recorder.mimeType}).arrayBuffer().then(this.complete.bind(this,e),this.completion.reject.bind(this.completion))};complete=(e,t)=>{if(this.signal.aborted){this.completion.reject(H());return}this.completion.resolve({bytes:t,...e})};async startPauseDetection(){let e=new AudioContext({sampleRate:x});this.audioContext=e;let t=e.resume();if(this.detector=await this.createDetector(),this.signal.throwIfAborted(),e.sampleRate!==16e3)throw Error(`Speech detection requires 16 kHz audio.`);await e.audioWorklet.addModule(a(`voice-activity-worklet.js`)),this.signal.throwIfAborted();let n=new AudioWorkletNode(e,`voice-activity-frames`);this.worklet=n,n.port.onmessage=this.receiveFrame,e.createMediaStreamSource(this.stream).connect(n),n.connect(e.destination),await t}receiveFrame=e=>{if(!(e.data instanceof Float32Array)||e.data.length!==512){this.failed();return}if(this.closed||this.recorder.state!==`recording`){e.data.fill(0);return}if(++this.queuedFrames>32){e.data.fill(0),this.failed();return}this.pending=this.pending.then(this.processFrame.bind(this,e.data)).catch(this.detectorFailed)};async processFrame(e){let t=this.detector;try{if(this.closed||!t)return;let n=await t.probability(e);if(this.closed){t.clear();return}let r=0;for(let t of e)r+=t*t;let i=Math.sqrt(r/e.length),a=n>=.5&&i>=R;this.updateActivity(a),this.onCapture({kind:`activity`,rms:i,speech:a})}finally{e.fill(0),--this.queuedFrames}}updateActivity(e){let t=performance.now();switch(this.activity.kind){case`waiting`:{let n=e?this.activity.activeMs+32:0;this.activity=n>=100?{kind:`speaking`,lastSoundAtMs:t,voicedMs:n}:{kind:`waiting`,activeMs:n};return}case`speaking`:{let{lastSoundAtMs:n,voicedMs:r}=this.activity;if(e){this.activity={kind:`speaking`,lastSoundAtMs:t,voicedMs:r+32};return}t-n>=v(r,this.expectedSpeechMs)&&this.stop(`pause`);return}default:return X(this.activity)}}stopPauseDetection=()=>{this.closed=!0,this.worklet&&(this.worklet.port.onmessage=null,this.worklet.port.close(),this.worklet.disconnect()),this.worklet=null,this.detector?.clear(),this.detector=null;let e=this.audioContext;this.audioContext=null,e&&e.close().catch(J)};stopTracks(){for(let e of this.stream.getTracks())e.stop()}cleanup(){for(let e of this.stream.getAudioTracks())e.removeEventListener(`ended`,this.interrupted),e.removeEventListener(`mute`,this.interrupted);this.stopPauseDetection(),this.stopTracks(),this.timer!==null&&clearTimeout(this.timer),this.signal.removeEventListener(`abort`,this.cancelled),this.recorder.removeEventListener(`dataavailable`,this.data),this.recorder.removeEventListener(`stop`,this.stopped),this.recorder.removeEventListener(`error`,this.failed)}};async function K(e,t,n){let r=-1,i=0,a=0,o=0,s=0,c=0,l=0,u=new Float32Array(512);try{for(let d=0;d<e.length;d+=u.length){t.throwIfAborted(),u.fill(0);let f=Math.min(u.length,e.length-d);u.set(e.subarray(d,d+f));let p=0;for(let e of u)p+=e*e;let m=Math.sqrt(p/f);if(o=Math.max(o,m),!(await n.probability(u)>=.5&&m>=R)){s+=p,c+=f;continue}r<0&&(r=d),i=d+f,a+=f,l+=p}}finally{u.fill(0),n.clear()}let d=Math.sqrt(s/Math.max(1,c)),f=Math.sqrt(l/Math.max(1,a)),p=a&&c?20*Math.log10(Math.max(f,1e-8)/Math.max(d,1e-8)):null,m={backgroundRms:d,speechToBackgroundDb:p,noise:d>=R&&(p===null||p<10)?`high`:c?`normal`:`unknown`},h=a/x*1e3;return r<0||h<600?{kind:`insufficient_audio`,voicedMs:h,peakFrameRms:o,...m}:{kind:`captured`,first:r,last:i,voicedMs:h,peakFrameRms:o,...m}}function q(e){for(let t=0;t<e.numberOfChannels;t+=1)e.getChannelData(t).fill(0)}function J(){}function Y(){let e,t;return{promise:new Promise((n,r)=>{e=n,t=r}),resolve:e,reject:t}}function X(e){throw Error(`Unexpected speech activity: ${String(e)}`)}export{i as C,n as S,m as _,I as a,f as b,M as c,x as d,C as f,g,h,B as i,N as l,y as m,W as n,F as o,v as p,V as r,A as s,L as t,S as u,p as v,r as x,s as y};
+const e={assetBase:null,microphonePreference:null,styleNonce:null,label:`Voice capture`,place:`this page`,site:`this site`,recoveryHint:`Reload the page and try again.`,cacheName:`seams-voice-capture-models-v1`};let t=e;function n(n){let r=new URL(n.assetBase,globalThis.document?.baseURI);r.pathname.endsWith(`/`)||(r.pathname+=`/`),t={assetBase:r,microphonePreference:n.microphonePreference??null,styleNonce:n.styleNonce??null,label:n.label??e.label,place:n.place??e.place,site:n.site??e.site,recoveryHint:n.recoveryHint??e.recoveryHint,cacheName:n.cacheName??e.cacheName}}function r(){return t}function i(){let e=t.styleNonce;return(typeof e==`function`?e():e)??void 0}function a(e){if(!t.assetBase)throw Error(`Call configureVoiceCapture({ assetBase }) before you load a voice asset.`);return new URL(e,t.assetBase)}const o=3e4;var s=class extends Error{constructor(e,t,n){super(t,n),this.kind=e,this.name=`VoiceAssetLoadError`}};async function c(){try{return await caches.open(r().cacheName)}catch{return null}}var l=class{pending=null;progress={kind:`checking_cache`};listeners=new Set;constructor(e){this.asset=e}get size(){return this.asset.bytes}async download(e){e&&(this.listeners.add(e),e(this.progress));try{this.pending||=this.bytes().catch(this.failed);let e=await this.pending;return this.publish({kind:`ready`}),e}finally{e&&this.listeners.delete(e)}}failed=e=>{throw this.pending=null,this.progress={kind:`checking_cache`},e};publish(e){this.progress=e;for(let t of this.listeners)t(e)}async valid(e){if(e.byteLength!==this.asset.bytes)return!1;let t=new Uint8Array(await crypto.subtle.digest(`SHA-256`,e)),n=``;for(let e of t)n+=e.toString(16).padStart(2,`0`);return n===this.asset.sha256}async bytes(){let{fileName:e,sha256:t}=this.asset;this.publish({kind:`checking_cache`});let n=await c(),r=new URL(`${e}?sha256=${t}`,a(``)).href;try{let e=await n?.match(r);if(e){let t=await e.arrayBuffer();if(await this.valid(t))return t;await n?.delete(r)}}catch{}let i;try{i=await this.fetchBytes()}catch(e){if(!(e instanceof s)||e.kind!==`connection`)throw e;i=await this.fetchBytes()}if(n)try{await n.put(r,new Response(i,{headers:{"Content-Type":`application/octet-stream`}}))}catch{}return i}async fetchBytes(){let{fileName:e}=this.asset,t=typeof this.asset.label==`function`?this.asset.label():this.asset.label,n=new AbortController,i=setTimeout(u.bind(null,n),o);this.publish({kind:`downloading`,received:0});try{let c=await fetch(a(e),{credentials:`omit`,cache:`reload`,signal:n.signal});if(!c.ok||!c.body){let e=[408,429,500,502,503,504].includes(c.status);throw new s(e?`connection`:`unavailable`,e?`${t} download was interrupted. Check your connection and retry.`:`${t} is unavailable from ${r().site}. ${r().recoveryHint}`)}let l=c.body.getReader(),f=new Uint8Array(this.asset.bytes),p=0;try{for(;;){let e=await l.read();if(e.done)break;if(clearTimeout(i),i=setTimeout(u.bind(null,n),o),p+e.value.length>f.length)throw new s(`integrity`,`${t} size is invalid. Retry loading.`);f.set(e.value,p),p+=e.value.length,this.publish({kind:`downloading`,received:p})}}finally{await l.cancel().catch(d),l.releaseLock()}if(p!==f.length||!await this.valid(f.buffer))throw new s(`integrity`,`${t} integrity check failed. Retry loading.`);return f.buffer}catch(e){throw e instanceof s?e:new s(`connection`,`${t} download was interrupted. Check your connection and retry.`,{cause:e})}finally{clearTimeout(i),n.abort()}}};function u(e){e.abort()}function d(){}const f=new l({fileName:`ort-wasm-simd-threaded.wasm`,sha256:`3398c10d07d229bd91b364548e130e0e51a8e5704b88c7c083ebbeb78842dee2`,bytes:14239897,label:()=>`${r().label} engine`});var p=class{asset;session=null;constructor(e){this.asset=new l(e)}prepare(){return this.session||=this.load().catch(this.failed),this.session}async run(e){let t=await this.prepare(),{Tensor:n}=await import(`./ort.wasm.bundle.min-CWqlff_l.js`),r=new n(`float32`,e,[1,e.length]);try{return await t.run({samples:r})}finally{r.dispose()}}failed=e=>{throw this.session=null,e};async load(){try{let e=await import(`./ort.wasm.bundle.min-CWqlff_l.js`);e.env.wasm.numThreads=1,e.env.wasm.wasmPaths=a(``).href,e.env.wasm.wasmBinary=await f.download();let t=await this.asset.download();return await e.InferenceSession.create(t,{executionProviders:[`wasm`]})}catch(e){if(e instanceof s)throw e;let{label:t,recoveryHint:n}=r();throw console.error(`[${t}] inference engine could not start`,e),new s(`initialization`,`${t} could not start. Retry loading. If it still fails, ${n.charAt(0).toLowerCase()}${n.slice(1)}`,{cause:e})}}};const m=new p({fileName:`silero-vad.onnx`,sha256:`2623a2953f6ff3d2c1e61740c6cdb7168133479b267dfef114a4a3cc5bdd788f`,bytes:2327524,label:`Speech detection`});async function h(){return new _(await m.prepare(),await import(`./ort.wasm.bundle.min-CWqlff_l.js`))}const g={probability:async()=>1,clear:()=>{}};var _=class{state=new Float32Array(256);context=new Float32Array(64);constructor(e,t){this.session=e,this.ort=t}async probability(e){if(e.length!==512)throw Error(`Speech detection requires a 512-sample frame.`);let t=new Float32Array(576);t.set(this.context),t.set(e,64);let n=new this.ort.Tensor(`float32`,t,[1,576]),r=new this.ort.Tensor(`float32`,this.state,[2,1,128]),i=new this.ort.Tensor(`int64`,BigInt64Array.of(16000n),[]);try{let t=await this.session.run({input:n,state:r,sr:i});try{let n=t.output?.data,r=t.stateN?.data;if(!(n instanceof Float32Array)||n.length!==1||!Number.isFinite(n[0])||n[0]<0||n[0]>1||!(r instanceof Float32Array)||r.length!==256)throw Error(`Speech detection returned an invalid result.`);return this.state.set(r),this.context.set(e.subarray(448)),n[0]}finally{for(let e of Object.values(t))e.dispose()}}finally{n.dispose(),r.dispose(),i.dispose(),t.fill(0)}}clear(){this.state.fill(0),this.context.fill(0)}};function v(e,t){return t!==null&&e>=t?800:1500}function y(e){return e.trim().replace(/[.!?,;:]+$/,``).split(/\s+/).filter(Boolean).reduce((e,t)=>e+b(t),0)*200}function b(e){let t=e.match(/[A-Za-z]+/g)?.length??0,n=e.match(/[0-9]/g)?.length??0,r=e.match(/[.\-_@]/g)?.length??0;return Math.max(1,t+n+r)}const x=16e3,S=8*x;function C(e){let t=Math.min(e.length,S),n=new Uint8Array(44+t*2),r=new DataView(n.buffer);w(n,0,`RIFF`),r.setUint32(4,36+t*2,!0),w(n,8,`WAVEfmt `),r.setUint32(16,16,!0),r.setUint16(20,1,!0),r.setUint16(22,1,!0),r.setUint32(24,x,!0),r.setUint32(28,x*2,!0),r.setUint16(32,2,!0),r.setUint16(34,16,!0),w(n,36,`data`),r.setUint32(40,t*2,!0);for(let n=0;n<t;n+=1){let t=Math.max(-1,Math.min(1,e[n]??0));r.setInt16(44+n*2,Math.round(t<0?t*32768:t*32767),!0)}return n}function w(e,t,n){for(let r=0;r<n.length;r+=1)e[t+r]=n.charCodeAt(r)}let T={kind:`unread`};function E(e){return typeof e==`string`&&e.length>0&&e!==`default`&&e!==`communications`}async function D(){if(T.kind!==`unread`)return T;let e={kind:`system_default`};try{let t=await r().microphonePreference?.read();if(typeof t==`object`&&t&&`kind`in t&&t.kind===`device`){let n=`deviceId`in t?t.deviceId:null;E(n)&&(e={kind:`device`,deviceId:n})}}catch{}return T.kind===`unread`&&(T=e),T}function O(e){let t={channelCount:1,echoCancellation:!0,noiseSuppression:!0,autoGainControl:!0};return e.kind===`device`&&(t.deviceId={exact:e.deviceId}),{audio:t,video:!1}}async function k(e){if(E(e.deviceId))return{kind:`device`,deviceId:e.deviceId};if(e.groupId)try{let t=await navigator.mediaDevices.enumerateDevices();for(let n of t)if(n.kind===`audioinput`&&n.groupId===e.groupId&&E(n.deviceId))return{kind:`device`,deviceId:n.deviceId}}catch{}return{kind:`system_default`}}async function A(e){e.throwIfAborted();let t=await D();e.throwIfAborted();let n,r=t.kind===`device`?`preferred`:`default`;try{n=await navigator.mediaDevices.getUserMedia(O(t))}catch(i){if(e.throwIfAborted(),t.kind!==`device`||!(i instanceof DOMException)||i.name!==`OverconstrainedError`&&i.name!==`NotFoundError`)throw i;n=await navigator.mediaDevices.getUserMedia(O({kind:`system_default`})),r=`replacement`}let i=j.bind(null,n);e.addEventListener(`abort`,i,{once:!0});try{e.throwIfAborted();let t=n.getAudioTracks()[0];if(!t||t.readyState!==`live`)throw new F;let i=t.getSettings(),a=await k(i);if(e.throwIfAborted(),t.readyState!==`live`)throw new F;return{stream:n,preference:a,selection:r,label:t.label||`Default microphone`,processing:{noiseSuppression:i.noiseSuppression??null,echoCancellation:i.echoCancellation??null,autoGainControl:i.autoGainControl??null,sampleRate:i.sampleRate??null}}}catch(e){throw i(),e}finally{e.removeEventListener(`abort`,i)}}function j(e){for(let t of e.getTracks())t.stop()}async function M(e){e.preference.kind!==`device`||e.selection===`replacement`||await P(e.preference)}async function N(){await P({kind:`system_default`})}async function P(e){T=e;try{await r().microphonePreference?.save(e)}catch{}}var F=class extends Error{constructor(){super(`The microphone disconnected or stopped. Check your microphone and try again.`),this.name=`VoiceMicrophoneInterruptedError`}};function I(e){return e.noise===`high`?`Background noise is high. Move closer to the microphone and try again.`:`Audio too brief or quiet: ${(e.voicedMs/1e3).toFixed(2)} of ${(e.requiredVoicedMs/1e3).toFixed(2)} seconds detected.`}var L=class extends Error{constructor(e){super(I(e)),this.quality=e,this.name=`InsufficientVoiceAudioError`}};const R=.006,z=8e3;function B(){if(V(),!navigator.userActivation.isActive){let{label:e,place:t}=r();throw Error(`Open ${t} and select Record to use ${e}.`)}}function V(){if(document.visibilityState!==`visible`||!document.hasFocus()){let{label:e,place:t}=r();throw Error(`Keep ${t} open and focused to use ${e}.`)}}var H=class extends Error{constructor(){super(`Select Record to use ${r().label}.`),this.name=`VoiceUserActionNeededError`}};function U(e){return new Promise(t=>setTimeout(t,e))}function W(){return Error(`${r().label} recording was cancelled.`)}function G(e){return e.detector===`level`?async()=>g:h}async function K(e,t,n={}){let r=n.expectedSpeechMs??null,i=G(n);if(e.aborted)throw W();let a=crypto.randomUUID();t({kind:`opening`});let o=await A(e),{stream:s}=o;try{if(e.aborted)throw W();let n=await new q(s,e,t,r,i).result.finally(t.bind(null,{kind:`processing`}));if(e.aborted)throw W();let c=await new OfflineAudioContext(1,x,x).decodeAudioData(n.bytes),l=new OfflineAudioContext(1,Math.ceil(c.duration*x),x),u=l.createBufferSource();u.buffer=c,u.connect(l.destination),u.start();try{let r=await l.startRendering();try{if(e.aborted)throw W();let s=r.getChannelData(0),c=await J(s,e,await i()),l={attemptId:a,capturedAtMs:Date.now(),durationMs:r.duration*1e3,recordingMs:n.recordingMs,endedBy:n.endedBy,voicedMs:c.voicedMs,peakFrameRms:c.peakFrameRms,levelThreshold:R,requiredVoicedMs:600,microphone:o.label,microphoneSelection:o.selection,processing:o.processing,backgroundRms:c.backgroundRms,speechToBackgroundDb:c.speechToBackgroundDb,noise:c.noise};if(t({kind:`measured`,quality:l}),c.kind===`insufficient_audio`)return{kind:`insufficient_audio`,quality:l};let u=s.slice(Math.max(0,c.first-3200),Math.min(s.length,c.last+3200));return await M(o),e.aborted&&(u.fill(0),e.throwIfAborted()),{kind:`captured`,samples:u,quality:l,lastSoundAtMs:n.lastSoundAtMs}}finally{Y(r)}}finally{Y(c)}}finally{for(let e of s.getTracks())e.stop()}}var q=class{recorder;chunks=[];timer=null;completion=Z();startedAtMs=performance.now();ended=null;activity={kind:`waiting`,activeMs:0};audioContext=null;worklet=null;detector=null;pending=Promise.resolve();queuedFrames=0;closed=!1;result=this.completion.promise;constructor(e,t,n,r,i){this.stream=e,this.signal=t,this.onCapture=n,this.expectedSpeechMs=r,this.createDetector=i,this.recorder=new MediaRecorder(e),this.recorder.addEventListener(`dataavailable`,this.data),this.recorder.addEventListener(`stop`,this.stopped),this.recorder.addEventListener(`error`,this.failed);for(let t of e.getAudioTracks())t.addEventListener(`ended`,this.interrupted),t.addEventListener(`mute`,this.interrupted);t.addEventListener(`abort`,this.cancelled,{once:!0}),this.timer=setTimeout(this.startupFailed,z),this.start()}async start(){try{if(await this.startPauseDetection(),this.signal.throwIfAborted(),this.startedAtMs=performance.now(),this.closed)return;this.recorder.start(),this.timer!==null&&clearTimeout(this.timer),this.timer=setTimeout(this.stop.bind(this,`limit`),z),this.onCapture({kind:`listening`})}catch(e){this.cleanup(),this.completion.reject(e instanceof Error?e:Error(`Speech detection could not start.`))}}data=e=>{this.chunks.push(e.data)};stop=e=>{this.recorder.state!==`inactive`&&(this.ended={endedBy:e,recordingMs:performance.now()-this.startedAtMs,lastSoundAtMs:this.activity.kind===`speaking`?this.activity.lastSoundAtMs:null},this.recorder.stop(),this.stopTracks())};cancelled=()=>{if(this.recorder.state===`inactive`){this.cleanup(),this.completion.reject(W());return}this.stop(`stream_ended`)};startupFailed=()=>{this.cleanup(),this.completion.reject(Error(`Speech detection startup timed out. Please try again.`))};detectorFailed=e=>{this.cleanup(),this.completion.reject(e instanceof Error?e:Error(`Speech detection failed.`))};failed=()=>{this.cleanup(),this.completion.reject(Error(`${r().label} could not record this microphone.`))};interrupted=()=>{this.cleanup(),this.completion.reject(new F)};stopped=()=>{let e=this.ended;if(this.cleanup(),this.signal.aborted){this.completion.reject(W());return}if(e===null){this.completion.reject(new F);return}new Blob(this.chunks,{type:this.recorder.mimeType}).arrayBuffer().then(this.complete.bind(this,e),this.completion.reject.bind(this.completion))};complete=(e,t)=>{if(this.signal.aborted){this.completion.reject(W());return}this.completion.resolve({bytes:t,...e})};async startPauseDetection(){let e=new AudioContext({sampleRate:x});this.audioContext=e;let t=e.resume();if(this.detector=await this.createDetector(),this.signal.throwIfAborted(),e.sampleRate!==16e3)throw Error(`Speech detection requires 16 kHz audio.`);await e.audioWorklet.addModule(a(`voice-activity-worklet.js`)),this.signal.throwIfAborted();let n=new AudioWorkletNode(e,`voice-activity-frames`);if(this.worklet=n,n.port.onmessage=this.receiveFrame,e.createMediaStreamSource(this.stream).connect(n),n.connect(e.destination),await Promise.race([t.catch(()=>{}),U(1e3)]),e.state!==`running`)throw new H}receiveFrame=e=>{if(!(e.data instanceof Float32Array)||e.data.length!==512){this.failed();return}if(this.closed||this.recorder.state!==`recording`){e.data.fill(0);return}if(++this.queuedFrames>32){e.data.fill(0),this.failed();return}this.pending=this.pending.then(this.processFrame.bind(this,e.data)).catch(this.detectorFailed)};async processFrame(e){let t=this.detector;try{if(this.closed||!t)return;let n=await t.probability(e);if(this.closed){t.clear();return}let r=0;for(let t of e)r+=t*t;let i=Math.sqrt(r/e.length),a=n>=.5&&i>=R;this.updateActivity(a),this.onCapture({kind:`activity`,rms:i,speech:a})}finally{e.fill(0),--this.queuedFrames}}updateActivity(e){let t=performance.now();switch(this.activity.kind){case`waiting`:{let n=e?this.activity.activeMs+32:0;this.activity=n>=100?{kind:`speaking`,lastSoundAtMs:t,voicedMs:n}:{kind:`waiting`,activeMs:n};return}case`speaking`:{let{lastSoundAtMs:n,voicedMs:r}=this.activity;if(e){this.activity={kind:`speaking`,lastSoundAtMs:t,voicedMs:r+32};return}t-n>=v(r,this.expectedSpeechMs)&&this.stop(`pause`);return}default:return Q(this.activity)}}stopPauseDetection=()=>{this.closed=!0,this.worklet&&(this.worklet.port.onmessage=null,this.worklet.port.close(),this.worklet.disconnect()),this.worklet=null,this.detector?.clear(),this.detector=null;let e=this.audioContext;this.audioContext=null,e&&e.close().catch(X)};stopTracks(){for(let e of this.stream.getTracks())e.stop()}cleanup(){for(let e of this.stream.getAudioTracks())e.removeEventListener(`ended`,this.interrupted),e.removeEventListener(`mute`,this.interrupted);this.stopPauseDetection(),this.stopTracks(),this.timer!==null&&clearTimeout(this.timer),this.signal.removeEventListener(`abort`,this.cancelled),this.recorder.removeEventListener(`dataavailable`,this.data),this.recorder.removeEventListener(`stop`,this.stopped),this.recorder.removeEventListener(`error`,this.failed)}};async function J(e,t,n){let r=-1,i=0,a=0,o=0,s=0,c=0,l=0,u=new Float32Array(512);try{for(let d=0;d<e.length;d+=u.length){t.throwIfAborted(),u.fill(0);let f=Math.min(u.length,e.length-d);u.set(e.subarray(d,d+f));let p=0;for(let e of u)p+=e*e;let m=Math.sqrt(p/f);if(o=Math.max(o,m),!(await n.probability(u)>=.5&&m>=R)){s+=p,c+=f;continue}r<0&&(r=d),i=d+f,a+=f,l+=p}}finally{u.fill(0),n.clear()}let d=Math.sqrt(s/Math.max(1,c)),f=Math.sqrt(l/Math.max(1,a)),p=a&&c?20*Math.log10(Math.max(f,1e-8)/Math.max(d,1e-8)):null,m={backgroundRms:d,speechToBackgroundDb:p,noise:d>=R&&(p===null||p<10)?`high`:c?`normal`:`unknown`},h=a/x*1e3;return r<0||h<600?{kind:`insufficient_audio`,voicedMs:h,peakFrameRms:o,...m}:{kind:`captured`,first:r,last:i,voicedMs:h,peakFrameRms:o,...m}}function Y(e){for(let t=0;t<e.numberOfChannels;t+=1)e.getChannelData(t).fill(0)}function X(){}function Z(){let e,t;return{promise:new Promise((n,r)=>{e=n,t=r}),resolve:e,reject:t}}function Q(e){throw Error(`Unexpected speech activity: ${String(e)}`)}export{n as C,r as S,g as _,B as a,s as b,A as c,S as d,x as f,h as g,y as h,V as i,M as l,v as m,H as n,I as o,C as p,K as r,F as s,L as t,N as u,m as v,i as w,f as x,p as y};

package/dist/browser/index.js

@@ -1 +1 @@
-import{S as e,_ as t,a as n,b as r,c as i,d as a,f as o,g as s,h as c,i as l,l as u,m as d,n as f,o as p,p as m,r as h,s as g,t as _,u as v,v as y,y as b}from"./chunks/microphone-CU5gQG6w.js";export{y as BrowserInferenceModel,_ as InsufficientVoiceAudioError,v as MAX_SAMPLES,a as SAMPLE_RATE,b as VoiceAssetLoadError,p as VoiceMicrophoneInterruptedError,e as configureVoiceCapture,c as createVoiceActivity,m as endPauseMs,d as expectedSpeechMs,s as levelDetector,g as openVoiceMicrophone,o as pcm16Wav,f as recordVoice,i as rememberVoiceMicrophone,h as requireVoiceForeground,l as requireVoiceUserAction,u as useSystemVoiceMicrophone,t as voiceActivityModel,r as voiceInferenceEngine,n as voiceQualityMessage};
+import{C as e,_ as t,a as n,b as r,c as i,d as a,f as o,g as s,h as c,i as l,l as u,m as d,n as f,o as p,p as m,r as h,s as g,t as _,u as v,v as y,x as b,y as x}from"./chunks/microphone-BQVF_tVV.js";export{x as BrowserInferenceModel,_ as InsufficientVoiceAudioError,a as MAX_SAMPLES,o as SAMPLE_RATE,r as VoiceAssetLoadError,g as VoiceMicrophoneInterruptedError,f as VoiceUserActionNeededError,e as configureVoiceCapture,s as createVoiceActivity,d as endPauseMs,c as expectedSpeechMs,t as levelDetector,i as openVoiceMicrophone,m as pcm16Wav,h as recordVoice,u as rememberVoiceMicrophone,l as requireVoiceForeground,n as requireVoiceUserAction,v as useSystemVoiceMicrophone,y as voiceActivityModel,b as voiceInferenceEngine,p as voiceQualityMessage};

package/dist/esm/card/phraseCard.js

@@ -1,13 +1,10 @@
+import { ROW_EASE } from "./rowMotion.js";
 //#region src/card/phraseCard.ts
 const SVG_NAMESPACE = "http://www.w3.org/2000/svg";
 const WIDTH = 308;
 const MID = 32;
 const SEGMENTS = 6;
 const GAP = 8;
-const FADE_OUT_MS = 150;
-const FADE_IN_MS = 300;
-const FADE_IN_DELAY_MS = 60;
-const ROW_STAGGER_MS = 40;
 const RIBBON_MOVE_MS = 180;
 const FLATTEN_MS = 200;
 const OPEN_MS = 200;
@@ -217,7 +214,7 @@ var PhraseCard = class {
 			["action", prompt.actionRow]
 		].filter(([key]) => this.rows[key] !== next[key] || key === "head" && this.rows.headline !== next.headline).map(([, element]) => element);
 		prompt.phrase.classList.toggle("seams-voiceid-phrase-dim", next.dim);
-		if (changed.length > 0 && !this.motion.matches) await Promise.all(changed.map((element) => fade(element, FADE_OUT_MS)));
+		if (changed.length > 0 && !this.motion.matches) await Promise.all(changed.map((element) => fade(element, 150)));
 		if (this.stale(version)) return;
 		const ribbonTop = prompt.ribbon.getBoundingClientRect().top;
 		const heightBefore = prompt.element.getBoundingClientRect().height;
@@ -232,7 +229,7 @@ var PhraseCard = class {
 			duration: RIBBON_MOVE_MS,
 			easing: EASE_MOVE
 		});
-		for (const [index, element] of changed.entries()) fadeIn(element, this.motion.matches, FADE_IN_DELAY_MS + index * ROW_STAGGER_MS);
+		for (const [index, element] of changed.entries()) fadeIn(element, this.motion.matches, 20 + index * 30);
 	}
 	markedText(text) {
 		if (typeof text === "string") return [text];
@@ -447,8 +444,8 @@ function fadeIn(element, reduced, delay) {
 		transform: "none"
 	}], {
 		delay: reduced ? 0 : delay,
-		duration: reduced ? 150 : FADE_IN_MS,
-		easing: EASE_IN_OUT,
+		duration: reduced ? 150 : 200,
+		easing: ROW_EASE,
 		fill: "both"
 	});
 }

package/dist/esm/card/prompt.js

@@ -1,5 +1,6 @@
 import { styleNonce } from "../config.js";
 import { useSystemVoiceMicrophone } from "../microphoneInput.js";
+import { ROW_EASE } from "./rowMotion.js";
 import { t as VoiceVisual } from "../chunks/visual-Di3ojTED.js";
 //#region src/card/prompt.ts
 const CSS = `
@@ -60,7 +61,7 @@ const CSS = `
 .seams-voiceid-prompt > .seams-voiceid-status { grid-area: 4 / 1; }
 .seams-voiceid-progress-slot { grid-row: 5; grid-column: 1; min-width: 0; align-self: center; }
 .seams-voiceid-progress-slot:empty { visibility: hidden; }
-.seams-voiceid-activation-surface .seams-voiceid-record-slot { transition: opacity 300ms cubic-bezier(0.2, 0, 0, 1) 140ms, visibility 0s linear 0s; }
+.seams-voiceid-activation-surface .seams-voiceid-record-slot { transition: opacity 200ms cubic-bezier(0.2, 0, 0, 1) 80ms, visibility 0s linear 0s; }
 .seams-voiceid-activation-surface[data-preparation="loading"] .seams-voiceid-record-slot { opacity: 0; visibility: hidden; transition: none; }
 @media (prefers-reduced-motion: reduce) {
   .seams-voiceid-prompt { transform: none; transition: opacity 150ms ease; }
@@ -316,8 +317,9 @@ const CSS = `
   font-weight: 500;
   border: 0;
   border-radius: var(--seams-shape-control, 10px);
-  background: var(--seams-colors-textPrimary, #eff2f3);
-  color: var(--seams-colors-colorBackground, #192024);
+  /* The theme's main button. A page that sets no button colours keeps the ink button. */
+  background: var(--seams-colors-buttonBackground, var(--seams-colors-textPrimary, #eff2f3));
+  color: var(--seams-colors-textButton, var(--seams-colors-colorBackground, #192024));
 }
 .seams-voiceid-prompt .seams-voiceid-primary {
   transition: opacity 180ms ease, visibility 0s linear 0s;
@@ -576,23 +578,7 @@ var VoicePrompt = class {
 		this.record.type = "button";
 		this.record.className = "seams-voiceid-primary";
 		this.record.textContent = "Record";
-		this.cancel = document.createElement("button");
-		this.cancel.type = "button";
-		this.cancel.className = "seams-voiceid-close";
-		const closeIcon = document.createElementNS("http://www.w3.org/2000/svg", "svg");
-		closeIcon.setAttribute("viewBox", "0 0 24 24");
-		closeIcon.setAttribute("width", "20");
-		closeIcon.setAttribute("height", "20");
-		closeIcon.setAttribute("aria-hidden", "true");
-		const closePath = document.createElementNS("http://www.w3.org/2000/svg", "path");
-		closePath.setAttribute("d", "M6 6l12 12M18 6L6 18");
-		closePath.setAttribute("fill", "none");
-		closePath.setAttribute("stroke", "currentColor");
-		closePath.setAttribute("stroke-width", "1.75");
-		closePath.setAttribute("stroke-linecap", "round");
-		closeIcon.append(closePath);
-		this.cancel.append(closeIcon);
-		this.cancel.setAttribute("aria-label", "Cancel");
+		this.cancel = cornerButton(document, "Cancel", CLOSE_PATH);
 		this.progress = document.createElement("div");
 		this.progress.className = "seams-voiceid-progress-slot";
 		const slot = document.createElement("div");
@@ -610,13 +596,22 @@ var VoicePrompt = class {
 		this.sizes.observe(this.element);
 	}
 	/**
-	* In an app's dialog with a step before this card, the corner control returns to that step.
-	* It still ends the card as cancelled; the Wallet tells the app that the person went back.
+	* In an app's dialog with a step before this card, Back in the leading corner returns to that
+	* step. With `close`, Close stays in the trailing corner and ends the whole dialog; without it,
+	* Back takes Close's place. Each one ends the card as cancelled. The result is the Back control.
 	*/
-	showBack() {
+	showBack(close) {
+		if (close) {
+			this.cancel.setAttribute("aria-label", "Close");
+			const back = cornerButton(this.element.ownerDocument, "Back", BACK_PATH);
+			back.classList.add("seams-voiceid-back");
+			this.cancel.before(back);
+			return back;
+		}
 		this.cancel.classList.add("seams-voiceid-back");
 		this.cancel.setAttribute("aria-label", "Back");
-		this.cancel.querySelector("path")?.setAttribute("d", "m15 18-6-6 6-6");
+		this.cancel.querySelector("path")?.setAttribute("d", BACK_PATH);
+		return this.cancel;
 	}
 	/**
 	* Moves the rows to a new state. The rows whose content changes fade out together, take their
@@ -973,11 +968,6 @@ function fadeIn(element, reduced, delay) {
 function assertNever(value) {
 	throw new Error(`Unexpected capture event: ${String(value)}`);
 }
-const ROW_OUT_MS = 150;
-const ROW_IN_MS = 300;
-const ROW_IN_DELAY_MS = 60;
-const ROW_STAGGER_MS = 40;
-const ROW_EASE = "cubic-bezier(0.2, 0, 0, 1)";
 const ROW_MOTION = "seams-voiceid-row";
 /** A frame that never reaches the card's size stops being followed after this time. */
 const FRAME_FILL_LIMIT_MS = 800;
@@ -1008,7 +998,7 @@ function fadeRowOut(row) {
 	const from = Number(getComputedStyle(row).opacity);
 	cancelRowMotion(row);
 	return row.animate([{ opacity: from }, { opacity: 0 }], {
-		duration: ROW_OUT_MS * from,
+		duration: 150 * from,
 		easing: "ease-in",
 		fill: "forwards",
 		id: ROW_MOTION
@@ -1028,8 +1018,8 @@ function fadeRowsIn(rows) {
 			opacity: to,
 			transform: "none"
 		}], {
-			delay: reduced ? 0 : ROW_IN_DELAY_MS + index * ROW_STAGGER_MS,
-			duration: reduced ? 150 : ROW_IN_MS,
+			delay: reduced ? 0 : 20 + index * 30,
+			duration: reduced ? 150 : 200,
 			easing: ROW_EASE,
 			fill: "backwards",
 			id: ROW_MOTION
@@ -1039,7 +1029,7 @@ function fadeRowsIn(rows) {
 function ignoreCancel() {}
 function continueFade({ row, opacity }) {
 	row.animate([{ opacity }, { opacity: 1 }], {
-		duration: ROW_IN_MS * (1 - opacity),
+		duration: 200 * (1 - opacity),
 		easing: ROW_EASE,
 		id: ROW_MOTION
 	});
@@ -1065,10 +1055,33 @@ function leaveSurface(element) {
 	element.style.pointerEvents = "none";
 	const from = Number(getComputedStyle(element).opacity);
 	element.animate([{ opacity: from }, { opacity: 0 }], {
-		duration: ROW_OUT_MS * from,
+		duration: 150 * from,
 		easing: "ease-in",
 		fill: "forwards"
 	}).finished.catch(ignoreCancel).finally(() => element.remove());
 }
+const CLOSE_PATH = "M6 6l12 12M18 6L6 18";
+const BACK_PATH = "m15 18-6-6 6-6";
+/** A control in a corner of the card, with an icon and no text. */
+function cornerButton(document, label, path) {
+	const button = document.createElement("button");
+	button.type = "button";
+	button.className = "seams-voiceid-close";
+	button.setAttribute("aria-label", label);
+	const icon = document.createElementNS("http://www.w3.org/2000/svg", "svg");
+	icon.setAttribute("viewBox", "0 0 24 24");
+	icon.setAttribute("width", "20");
+	icon.setAttribute("height", "20");
+	icon.setAttribute("aria-hidden", "true");
+	const line = document.createElementNS("http://www.w3.org/2000/svg", "path");
+	line.setAttribute("d", path);
+	line.setAttribute("fill", "none");
+	line.setAttribute("stroke", "currentColor");
+	line.setAttribute("stroke-width", "1.75");
+	line.setAttribute("stroke-linecap", "round");
+	icon.append(line);
+	button.append(icon);
+	return button;
+}
 //#endregion
 export { VoicePrompt, leaveSurface };

package/dist/esm/card/rowMotion.js

@@ -0,0 +1,8 @@
+//#region src/card/rowMotion.ts
+const ROW_OUT_MS = 150;
+const ROW_IN_MS = 200;
+const ROW_IN_DELAY_MS = 20;
+const ROW_STAGGER_MS = 30;
+const ROW_EASE = "cubic-bezier(0.2, 0, 0, 1)";
+//#endregion
+export { ROW_EASE, ROW_IN_DELAY_MS, ROW_IN_MS, ROW_OUT_MS, ROW_STAGGER_MS };

package/dist/esm/card/surface.js

@@ -2,7 +2,7 @@ import { captureConfig } from "../config.js";
 import { voiceActivityModel } from "../voiceActivity.js";
 import { expectedSpeechMs } from "../speechPause.js";
 import { VoiceMicrophoneInterruptedError } from "../microphoneInput.js";
-import { recordVoice, voiceQualityMessage } from "../microphone.js";
+import { VoiceUserActionNeededError, recordVoice, voiceQualityMessage } from "../microphone.js";
 import { PhraseCard } from "./phraseCard.js";
 import { VoicePrompt, leaveSurface } from "./prompt.js";
 //#region src/card/surface.ts
@@ -24,12 +24,17 @@ var CardSurface = class {
 	unmounted;
 	outer;
 	detector;
+	pressed;
+	/** Back, when the card shows it apart from Close. */
+	back;
 	recorder = null;
 	recording = null;
 	choice = null;
 	view = null;
 	started = false;
 	closed = false;
+	/** The next recording starts with no press. A card that continues another never does. */
+	autoStart;
 	/** The voiced time of the whole sentence, after which a shorter pause ends a recording. */
 	expectedSpeechMs;
 	/** The speech detector has started; before that, a recording waits for it to load. */
@@ -39,6 +44,7 @@ var CardSurface = class {
 		const outer = input.signal;
 		this.outer = outer;
 		this.detector = input.detector;
+		this.autoStart = input.autoStart === true;
 		this.expectedSpeechMs = expectedSpeechMs(sentence);
 		const handedOver = document.querySelector(".seams-voiceid-activation-surface[data-seams-voiceid-handoff]");
 		if (input.detector === "level") this.speechDetectorReady = true;
@@ -56,12 +62,16 @@ var CardSurface = class {
 		delete this.prompt.phrase.dataset.voiceidPhrase;
 		this.prompt.status.textContent = PROMPT_MESSAGE;
 		this.prompt.record.textContent = "Record";
-		if (frame.back) this.prompt.showBack();
+		const controls = frame.controls ?? "close";
+		this.pressed = frame.pressed;
+		const back = controls === "close" ? null : this.prompt.showBack(controls === "back_and_close");
+		this.back = back === this.prompt.cancel ? null : back;
 		const handedOverPrompt = handedOver?.querySelector(":scope > .seams-voiceid-prompt") ?? null;
 		const rows = handedOverPrompt ? handedOverRows(handedOverPrompt) : null;
 		if (handedOverPrompt) {
 			this.prompt.continueFrom(handedOverPrompt);
 			this.prompt.record.hidden = true;
+			this.autoStart = false;
 		}
 		this.card = new PhraseCard(this.prompt, {
 			label: PROMPT_LABEL,
@@ -71,7 +81,9 @@ var CardSurface = class {
 		}, rows);
 		this.element.append(this.prompt.element);
 		this.prompt.record.addEventListener("click", this.press);
-		this.prompt.cancel.addEventListener("click", this.cancel);
+		const corner = controls === "back" ? this.pressBack : this.pressClose;
+		this.prompt.cancel.addEventListener("click", corner);
+		this.back?.addEventListener("click", this.pressBack);
 		this.element.addEventListener("cancel", this.cancel);
 		outer.addEventListener("abort", this.cancel);
 		if (handedOver) handedOver.replaceWith(this.element);
@@ -91,7 +103,8 @@ var CardSurface = class {
 		if (this.view) {
 			this.view = null;
 			this.card.showPrompt();
-		}
+		} else if (this.autoStart) this.begin(recording);
+		this.autoStart = false;
 		return recording.promise;
 	}
 	show(view) {
@@ -113,7 +126,9 @@ var CardSurface = class {
 		this.outer.removeEventListener("abort", this.cancel);
 		this.element.removeEventListener("cancel", this.cancel);
 		this.prompt.record.removeEventListener("click", this.press);
-		this.prompt.cancel.removeEventListener("click", this.cancel);
+		this.prompt.cancel.removeEventListener("click", this.pressBack);
+		this.prompt.cancel.removeEventListener("click", this.pressClose);
+		this.back?.removeEventListener("click", this.pressBack);
 		this.card.dispose();
 		this.prompt.dispose();
 		leaveSurface(this.element);
@@ -125,12 +140,15 @@ var CardSurface = class {
 			return;
 		}
 		if (!this.recording || this.started) return;
+		this.begin(this.recording);
+	};
+	begin(recording) {
 		this.started = true;
 		this.prompt.record.disabled = true;
 		this.prompt.status.textContent = this.speechDetectorReady ? "Opening microphone…" : `Loading ${captureConfig().label}…`;
-		this.attempt(this.recording);
-	};
-	async attempt(recording) {
+		this.attempt(recording, this.autoStart);
+	}
+	async attempt(recording, unpressed = false) {
 		const recorder = new AbortController();
 		this.recorder = recorder;
 		try {
@@ -160,6 +178,10 @@ var CardSurface = class {
 				this.retry(error.message);
 				return;
 			}
+			if (unpressed && error instanceof VoiceUserActionNeededError) {
+				this.retry(PROMPT_MESSAGE, "Record");
+				return;
+			}
 			this.started = false;
 			this.recording = null;
 			recording.reject(error instanceof Error ? error : new Error(String(error)));
@@ -173,9 +195,9 @@ var CardSurface = class {
 		if (event.kind === "processing") this.card.beginCheck();
 		else this.prompt.capture(event);
 	};
-	retry(message) {
+	retry(message, action = "Try again") {
 		this.started = false;
-		this.prompt.record.textContent = "Try again";
+		this.prompt.record.textContent = action;
 		this.prompt.ready(message);
 		this.prompt.record.focus();
 	}
@@ -185,6 +207,13 @@ var CardSurface = class {
 		settle?.(choice);
 	}
 	/** A close during a check is not possible: the button is hidden while the caller works. */
+	pressBack = () => this.pressControl("back");
+	pressClose = () => this.pressControl("close");
+	pressControl(control) {
+		if (this.view?.kind === "check") return;
+		this.pressed?.(control);
+		this.cancel();
+	}
 	cancel = () => {
 		if (this.view?.kind === "check" && !this.outer.aborted) return;
 		if (this.recording) {

package/dist/esm/index.js

@@ -4,5 +4,5 @@ import { createVoiceActivity, levelDetector, voiceActivityModel } from "./voiceA
 import { endPauseMs, expectedSpeechMs } from "./speechPause.js";
 import { MAX_SAMPLES, SAMPLE_RATE, pcm16Wav } from "./wav.js";
 import { VoiceMicrophoneInterruptedError, openVoiceMicrophone, rememberVoiceMicrophone, useSystemVoiceMicrophone } from "./microphoneInput.js";
-import { InsufficientVoiceAudioError, recordVoice, requireVoiceForeground, requireVoiceUserAction, voiceQualityMessage } from "./microphone.js";
-export { BrowserInferenceModel, InsufficientVoiceAudioError, MAX_SAMPLES, SAMPLE_RATE, VoiceAssetLoadError, VoiceMicrophoneInterruptedError, configureVoiceCapture, createVoiceActivity, endPauseMs, expectedSpeechMs, levelDetector, openVoiceMicrophone, pcm16Wav, recordVoice, rememberVoiceMicrophone, requireVoiceForeground, requireVoiceUserAction, useSystemVoiceMicrophone, voiceActivityModel, voiceInferenceEngine, voiceQualityMessage };
+import { InsufficientVoiceAudioError, VoiceUserActionNeededError, recordVoice, requireVoiceForeground, requireVoiceUserAction, voiceQualityMessage } from "./microphone.js";
+export { BrowserInferenceModel, InsufficientVoiceAudioError, MAX_SAMPLES, SAMPLE_RATE, VoiceAssetLoadError, VoiceMicrophoneInterruptedError, VoiceUserActionNeededError, configureVoiceCapture, createVoiceActivity, endPauseMs, expectedSpeechMs, levelDetector, openVoiceMicrophone, pcm16Wav, recordVoice, rememberVoiceMicrophone, requireVoiceForeground, requireVoiceUserAction, useSystemVoiceMicrophone, voiceActivityModel, voiceInferenceEngine, voiceQualityMessage };

package/dist/esm/microphone.js

@@ -36,6 +36,21 @@ function requireVoiceForeground() {
 		throw new Error(`Keep ${place} open and focused to use ${label}.`);
 	}
 }
+/**
+* The browser starts audio only after the person acts on the page (a click or a key). A
+* recording that starts by itself gets this error, and the caller then shows its Record control.
+*/
+var VoiceUserActionNeededError = class extends Error {
+	constructor() {
+		super(`Select Record to use ${captureConfig().label}.`);
+		this.name = "VoiceUserActionNeededError";
+	}
+};
+/** The time that a permitted audio context needs at most to start, after its setup. */
+const AUDIO_START_MS = 1e3;
+function elapsed(ms) {
+	return new Promise((resolve) => setTimeout(resolve, ms));
+}
 function cancelled() {
 	return /* @__PURE__ */ new Error(`${captureConfig().label} recording was cancelled.`);
 }
@@ -243,7 +258,8 @@ var VoiceRecording = class {
 		worklet.port.onmessage = this.receiveFrame;
 		context.createMediaStreamSource(this.stream).connect(worklet);
 		worklet.connect(context.destination);
-		await resumed;
+		await Promise.race([resumed.catch(() => {}), elapsed(AUDIO_START_MS)]);
+		if (context.state !== "running") throw new VoiceUserActionNeededError();
 	}
 	receiveFrame = (event) => {
 		if (!(event.data instanceof Float32Array) || event.data.length !== 512) {
@@ -425,4 +441,4 @@ function assertNeverActivity(activity) {
 	throw new Error(`Unexpected speech activity: ${String(activity)}`);
 }
 //#endregion
-export { InsufficientVoiceAudioError, recordVoice, requireVoiceForeground, requireVoiceUserAction, voiceQualityMessage };
+export { InsufficientVoiceAudioError, VoiceUserActionNeededError, recordVoice, requireVoiceForeground, requireVoiceUserAction, voiceQualityMessage };

package/dist/example/example.js

@@ -9,7 +9,7 @@ configureVoiceCapture({ assetBase: '../assets/' });
 const ENROLLMENT_PHRASES = [
   'My voice is my signature for this wallet.',
   'I send money only when I say so.',
-  'Seams, please learn how I sound today.',
+  'Please learn how I sound today.',
 ];
 
 const log = document.querySelector('#log');
@@ -18,7 +18,13 @@ const detector = () => (document.querySelector('#level').checked ? 'level' : 'mo
 
 /** Shows one sentence, records it and returns its WAV, or null when the person closed the card. */
 async function recordSentence(sentence, signal) {
-  const surface = openPhraseCard({ sentence, signal, detector: detector(), frame: { parent } });
+  const surface = openPhraseCard({
+    sentence,
+    signal,
+    detector: detector(),
+    autoStart: document.querySelector('#auto').checked,
+    frame: { parent },
+  });
   try {
     const recording = await surface.record();
     if (recording.kind === 'cancelled') return null;

package/dist/example/index.html

@@ -17,6 +17,7 @@
       <h1>voice-capture</h1>
       <p>This page uses only <code>@seams/voice-capture</code>. The audio stays in the page.</p>
       <label><input type="checkbox" id="level" /> Record with no model download (the detector <code>level</code>)</label>
+      <label><input type="checkbox" id="auto" /> Start the recording with no press on Record (<code>autoStart</code>)</label>
       <div class="row">
         <button class="start" id="sentence">Record one sentence</button>
         <button class="start" id="enroll">Record three enrollment phrases</button>

package/dist/types/card/index.d.ts

@@ -1,4 +1,4 @@
-export { openPhraseCard, type PhraseCardFrame, type PhraseCardInput } from './surface';
+export { openPhraseCard, type PhraseCardControl, type PhraseCardControls, type PhraseCardFrame, type PhraseCardInput, } from './surface';
 export type { PhraseCheckView, PhraseChoice, PhraseRecording, PhraseResultView, PhraseSegmentState, PhraseSurface, PhraseText, PhraseView, } from './types';
 export { leaveSurface, VoicePrompt, type PhraseDisplay, type PromptChange } from './prompt';
 export { PhraseCard, type HandedOverRows, type PromptCopy } from './phraseCard';

package/dist/types/card/prompt.d.ts

@@ -46,10 +46,11 @@ export declare class VoicePrompt {
     /** The text to show for a phrase, such as the phrase with its commas. */
     display?: PhraseDisplay);
     /**
-     * In an app's dialog with a step before this card, the corner control returns to that step.
-     * It still ends the card as cancelled; the Wallet tells the app that the person went back.
+     * In an app's dialog with a step before this card, Back in the leading corner returns to that
+     * step. With `close`, Close stays in the trailing corner and ends the whole dialog; without it,
+     * Back takes Close's place. Each one ends the card as cancelled. The result is the Back control.
      */
-    showBack(): void;
+    showBack(close: boolean): HTMLButtonElement;
     /**
      * Moves the rows to a new state. The rows whose content changes fade out together, take their
      * new content and fade in one after another; the other rows stay still. A change that comes

package/dist/types/card/rowMotion.d.ts

@@ -0,0 +1,5 @@
+export declare const ROW_OUT_MS = 150;
+export declare const ROW_IN_MS = 200;
+export declare const ROW_IN_DELAY_MS = 20;
+export declare const ROW_STAGGER_MS = 30;
+export declare const ROW_EASE = "cubic-bezier(0.2, 0, 0, 1)";

package/dist/types/card/surface.d.ts

@@ -6,19 +6,35 @@ export type PhraseCardFrame = Readonly<{
     parent?: HTMLElement;
     /** More class names for the surface element, such as the host's theme class. */
     className?: string;
-    /** Shows the Back control of the prompt. */
-    back?: boolean;
+    /**
+     * The corner controls of the card. `close` (default): Close in the trailing corner. `back`: Back
+     * in the leading corner, in place of Close. `back_and_close`: both. Each one, and Escape, ends
+     * the card as cancelled.
+     */
+    controls?: PhraseCardControls;
+    /** Runs when the person presses Back or Close, before the card ends. Escape is not a press. */
+    pressed?(control: PhraseCardControl): void;
     /** The host sizes its frame to the surface, so the card follows the frame. */
     hugged?: boolean;
     /** Runs when the surface is in the document. Its return value runs when the surface closes. */
     mounted?(element: HTMLElement): (() => void) | void;
 }>;
+/** The corner controls of a phrase card. */
+export type PhraseCardControls = 'close' | 'back' | 'back_and_close';
+/** One corner control of a phrase card. */
+export type PhraseCardControl = 'back' | 'close';
 export type PhraseCardInput = Readonly<{
     sentence: string;
     signal: AbortSignal;
     frame?: PhraseCardFrame;
     /** The speech detector of the recordings. Default: `model`. */
     detector?: VoiceRecordingOptions['detector'];
+    /**
+     * The first recording starts when the caller asks for it, with no press on Record. Use it
+     * when the person just asked for this card, with a click or a key. If the browser does not
+     * start audio without a press, the card stays with its Record button.
+     */
+    autoStart?: boolean;
 }>;
 /**
  * Shows the given sentence in the prompt. The surface records clips with the quality check, and

package/dist/types/index.d.ts

@@ -1,5 +1,5 @@
 export { configureVoiceCapture, type VoiceCaptureConfig, type VoiceMicrophonePreference, type VoiceMicrophonePreferenceStore, } from './config';
-export { recordVoice, requireVoiceForeground, requireVoiceUserAction, voiceQualityMessage, InsufficientVoiceAudioError, type VoiceActivityEvent, type VoiceCaptureEvent, type VoiceCaptureObserver, type VoiceCaptureQuality, type VoiceRecordingOptions, type VoiceRecordingResult, } from './microphone';
+export { recordVoice, requireVoiceForeground, requireVoiceUserAction, voiceQualityMessage, InsufficientVoiceAudioError, VoiceUserActionNeededError, type VoiceActivityEvent, type VoiceCaptureEvent, type VoiceCaptureObserver, type VoiceCaptureQuality, type VoiceRecordingOptions, type VoiceRecordingResult, } from './microphone';
 export { openVoiceMicrophone, rememberVoiceMicrophone, useSystemVoiceMicrophone, VoiceMicrophoneInterruptedError, type VoiceMicrophone, } from './microphoneInput';
 export { endPauseMs, expectedSpeechMs } from './speechPause';
 export { createVoiceActivity, levelDetector, voiceActivityModel, type SpeechDetector, } from './voiceActivity';

package/dist/types/microphone.d.ts

@@ -73,6 +73,13 @@ export type VoiceRecordingOptions = Readonly<{
 export declare function requireVoiceUserAction(): void;
 /** Throws unless the page is visible and has the focus. */
 export declare function requireVoiceForeground(): void;
+/**
+ * The browser starts audio only after the person acts on the page (a click or a key). A
+ * recording that starts by itself gets this error, and the caller then shows its Record control.
+ */
+export declare class VoiceUserActionNeededError extends Error {
+    constructor();
+}
 /**
  * Records one utterance at 16 kHz. The recording ends after a pause, or after 8 s. A recording
  * with less than 0.6 s of speech is `insufficient_audio`. The samples hold the speech with 0.2 s

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@seams/voice-capture",
-  "version": "0.67.1",
+  "version": "0.68.0",
   "type": "module",
   "description": "Records one spoken sentence in the browser, with speech detection, a quality check and a phrase card",
   "license": "Apache-2.0",