@pipeup/voice 0.8.2 → 0.9.1

package/README.md

@@ -24,15 +24,20 @@ mount();
 
 ## What it sends, and to whom
 
-Pipeup itself makes no request. **Your browser, not Pipeup, may send your voice to Google or Apple.**
+Pipeup itself makes no request. **Your browser, not Pipeup, may send your voice to its maker's speech service.**
 
 - The first time you press the microphone, Pipeup asks you, in one sentence, before anything starts:
-  - **Chrome, when it can work on your device:** "Your words are worked out on this device. Nothing is sent."
-  - **Chrome, when it needs a speech pack first:** the browser downloads the pack for your language from
-    Google, once, and your words then stay on your device.
-  - **Otherwise** (Chrome without a pack, and always in Safari): your voice goes to Google (in Chrome) or
-    Apple (in Safari) to be turned into words. Pipeup doesn't keep it.
-- **Safari can't tell Pipeup where it listens**, so Pipeup always shows the service sentence there.
+  - **When the browser can work on your device:** "Your words are worked out on this device. Nothing is sent."
+  - **When it needs a speech pack first:** the browser downloads the pack for your language from its maker,
+    once, and your words then stay on your device.
+  - **Otherwise** (Chrome without a pack): your voice goes to the browser's speech service to be turned into
+    words. Pipeup doesn't keep it.
+  - **Safari** may work some languages (such as English) out on the device and send others to Apple, and
+    can't tell Pipeup which, so its sentence says so: "Safari uses Apple's speech recognition. Depending on the language and
+    device the words are worked out here or your voice is sent to Apple; Pipeup can't tell which, and doesn't
+    keep it."
+  - **A language the engine can't do at all** (Firefox, for a language with no model) is said before anything
+    starts, with Start greyed until you choose another.
 - Your answer and the language are kept in this browser's local storage. Where the browser can't keep
   them, the panel says "This browser won't remember this choice". On a page opened from disk (`file://`) the
   choice may apply to every local file.
@@ -42,9 +47,18 @@ The package states this in `package.json` under `pipeup.network`; `pipeup check`
 ## Where it works
 
 - **Chrome and Edge:** `SpeechRecognition`, on device where the browser allows it.
-- **Safari:** `webkitSpeechRecognition`, through Apple's service.
-- **Firefox:** has no speech engine, so Pipeup draws no microphone at all. The system's own dictation
-  (macOS, Windows, iOS and Android all have one) still types into any field, including Pipeup's.
+- **Safari:** `webkitSpeechRecognition`, through Apple's speech recognition. Safari asks twice the first time:
+  for the microphone, and to "capture your audio and use it for speech recognition"; macOS may also ask once
+  whether Safari may use Speech Recognition. **A refusal lasts for the page**: allow it, then reload. If the
+  microphone is allowed but Pipeup says Safari isn't allowed to use Apple's speech recognition, turn on Siri or
+  Dictation, or allow Safari under System Settings › Privacy & Security › Speech Recognition, then reload.
+  Languages Apple sends to its service stop by themselves after about a minute; press the microphone again.
+  Safari can send everything said so far again; Pipeup writes it once, so a phrase repeated word for word
+  straight after itself is written once too.
+- **Firefox 157 to 159:** has no speech engine, so Pipeup draws no microphone at all and says so once in the
+  browser console. The system's own dictation (macOS, Windows, iOS and Android all have one) still types into
+  any field, including Pipeup's. Firefox 160 Beta and Nightly have an on-device engine (a one-time model
+  download from Mozilla); Mozilla intends to ship it in Firefox 161.
 
 On a page opened from disk, **Chrome and Safari ask for the microphone again on every load**; that is the
 browsers' rule, and Pipeup's own choice above isn't asked again.
@@ -56,7 +70,10 @@ browsers' rule, and Pipeup's own choice above isn't asked again.
 - Words in progress show in the box as you speak and become final as the engine settles on them.
 - The language is the nearest `lang` attribute above the commentable area, else the page's, else the
   browser's. The panel shows it and lets you change it.
-- Problems (microphone blocked, nothing heard, no network, no microphone) are said in plain words.
+- Problems (microphone blocked, nothing heard, no network, no microphone, the browser's speech service off)
+  are said in plain words, naming what to change; where the engine gives its own reason, it is passed on.
+  After any of them the microphone works again at once.
+- Words the engine delivers twice (Safari can send everything said so far again) are written once.
 - It announces "Listening" and "Stopped listening" to screen readers. Pipeup holds announcements while you
   are writing, so they may be heard only after you finish the comment.