Talk Forward Screen Reader

And so,

I understand how the three-finger rotor works.

You turn all three fingers clockwise or counterclockwise.

In addition, for either version of the rotor gestures to work easier, you may wish to unassign the two-finger groups actions and change them to something else. Or use the three-finger rotor for groups and keep the two-finger rotor gesture actions as they are by default. Or change the rotor actions to the two-finger left/right swipes.
At least on my phone with a screen protector, it works better that way.

As for the Talk Forward keyboard...

The only issue I have with it, is that it doesn't defult to, default. It defaults to characters. It should be defult, then you can cycle to characters if needed. Or if you're in an edit box, then I guess characters may may more since. But not outside of it. But then again, the enhanced keyboard with browserse mode uses arrorws going by letters and not by the default element. So, I'm not sure which is right.

Can't waitron try and make a voiceover keyboard layout since I found the scripting instructions. I never dealt with LUA before. Shouldn't' be too hard.

Almost forgot to mention, the practice screen currently doesn't have all talk forward bindings yet. I guess that's in the works?

• Stability, at depth: until now, a single failing feature could switch the whole screen reader off. A user reported Talk Forward restarting over and over on their Xiaomi after a reboot, before the first unlock — the cause was the translation feature crashing because its language models cannot start at that point. Rather than fix that one case, we went through everything that runs in the background or reacts to a gesture and put a net under it: a failure now costs the feature involved, never the reader. In practice that covers every gesture, every rotor entry, every menu, every accessibility event, and some thirty background tasks.
• Translation: when its models cannot start, it says so instead of stopping the reader.

• New, in beta: screen recognition. Some apps expose nothing usable — a game, a drawn view, an unlabeled button. Talk Forward can now rebuild what is on screen by combining what the app does reveal of its geometry with what text recognition reads from the pixels, then let you move through that model by swiping and activate by double-tapping. You switch it on app by app, from the "Screen recognition" rotor entry, and it stays inactive everywhere else. Text recognition runs offline and free for everyone; supporters additionally get AI analysis and a shared cache, which names silent icons once for every user. This is a first version: explore-by-touch is not available yet, and protected screens (banking, PIN entry) cannot be analyzed. Feedback welcome.

Also included since the previous stable release:

• Voice: fixed the reader going completely silent after a reboot when the speech engine (e.g. Vocalizer) failed to start — it now recovers automatically, falling back to eSpeak if needed.
• More robust startup: if an optional feature fails to start (device-dependent), the reader still comes up with at least speech and touch.
• Three-finger rotor: a new rotor alongside the two-finger one — the middle finger pivots while the outer two rotate around it.
• Field-aware letter keyboard: a dialpad for number fields and PINs, a number row for passwords, and @ and .com keys within reach for e-mail and web addresses.
• Calls: the answer/end-call gesture now works with WhatsApp, Telegram and Signal calls too, not just carrier calls.
• Physical keyboard shortcuts: the Alt key (left or right) can be used as the keymap modifier.
• Italian braille: updated comp6 table (Monza standard).
• Volume keys: fixed them sometimes feeling dead — they only move the cursor when the keyboard is actually up, otherwise they change the volume as before.
• New "Add to clipboard" action (assignable to a gesture): appends the focused item's text to whatever is already on the clipboard — fire it on several items to accumulate them.
• Braille keyboard: three-finger-up "send" no longer sometimes switches conversation instead of sending in some apps (e.g. Google Messages' web view).
• The braille/letter keyboard no longer triggers unwanted screen rotation.
• Updated for the latest Android and Google Play requirements.

It's a dilemma!

Is it better to become a Talk Supporter from the play Store or is it better to go the PayPal route. I myself can do either one of said mentioned options honestly. I've been just waiting for the right calling on wish one to go with.

Honestly?

Going the play store route means I get points for purchasing (particularly since I have Play Pass for an entire year).

So, the question then is:

If I have the play store version of Talk Forward on one phone, but the non-play store version on another phone, do I have to do a onetime payment for each phone?

Which I do not mind doing.

Ah, thanks for sharing!

Yeah, that does help.

, but it sure makes a difference when supporting a program, you use every day.

Especially since I adore the color gold!

, things are a lot smoother and that one issue with Samsung being odd that y'all mention hasn't cropped up on my S23+ as of now.
T-Mobile users got the patch on the 31st of July, not surprisingly.

As for the Screen Rec still being in beta, playing games such as Dice Dreams and Scattered Slots currently can't be done at the moment. Can't wait though for thy feature to be more well-rounded!
Using Prudence for now on the side.

Interestingly (in conjunction with describing the screen), the Visual Aid feature with Dice Dreams (at least ffor rolling the dice), works!

Although I'm not the dev, I shall try and help out best I can.

With that being said, the main question I have for you is:

Which phone are you using?

As for answers:

1. TF support is onetime payment.
2. I'll have someone else answer that one.
3. Adding voices to the rotor can be done from

Talk Forward Settings>Voice>Add A Voice.

Eeep! Must have missed that detail when first reading your prior post.
In that case,
As Talk Forward 26.8.1 is now available in Google Play, which version are you using? Non-Play Store version or Play Store.

As for TTS speeds when it comes to Samsung, I first wrote that you first have to crank up the normal Text to Speech setting all the way up to 100. (Referring to the system's settings).

But after forgetting that ESPeak has its own Rate settings that's inside its app, you may have to adjust it there.

i get the version of ESPeak NG from

https://github.com/espeak-ng/espeak-ng/releases/download/1.52.0/espeak-1.52.0-signed.apk

, because the ones on the play store (unless that has changed), are either out of date or have ads!

As for voice rates if using the rotor (particularly with ESpeak), choose the DeFault Voice and change its voice via its app.
Changing TTS settings for the rotor (at least for me), I always had to explore by touch to find the rate/pitch setting, double-tapping and holding till I got the speed/pitch right. Up/down swipes on the bar never worked for me for some reason.
Unless that's been fixed now?

You can now find (under the Getting Started & Support) heading of the main post, links to both the Talk Forward Help Center & Changelog.

In short, quite a lot!

Version 26.9.3

Released September 20, 2026Android Talk Forward 26.9.3 Offline dictation, with Parakeet • A new dictation engine runs entirely on the device and writes almost as well as the online smart one. Nothing is sent anywhere, it works with no network, it punctuates and capitalizes on its own, and it keeps the English words in the middle of a French sentence. On a Pixel 9, twenty-four seconds of speech come back in a second and a half; the text arrives when you lift your finger. • The model is not in the app: it is downloaded from Settings, Dictation, on the "Parakeet model" row. It is 478 MB, downloads over Wi-Fi only, and takes about 670 MB on the device. Progress is announced, you can leave the screen while it runs, and the same row later deletes it. Speech recognition by NVIDIA Parakeet TDT 0.6B v3 and sherpa-onnx. • Online dictation stays the default engine: it still makes slightly fewer mistakes when you hesitate. Dictation on the device • An "Offline only" engine forces the device's own recognizer instead of letting the system pick a service free to send your voice to a server. A device with no local model says so rather than quietly dictating over the network. • Spoken punctuation is back, and local: say "comma", "period", "new line", and the mark is inserted, with the typography the language wants. The setting is in Dictation and applies to every engine. • A gesture can now edit text by voice: "replace Saturday with Sunday", "delete the last three words", "capitalize Lyon", "read it back", "undo". The command is caught by the device's own recognizer, so it answers instantly, costs nothing and works offline, and every edit says what it changed. An utterance that matches no command changes nothing and is read back to you: it is never inserted into the field. • Numbers spoken in Bengali, Hindi, Tamil, Thai and other scripts are finally understood in those commands. • The Dictation screen is reorganized: the options that belong to the smart engine appear only while that engine is selected. On-screen keyboards • The letter keyboard can show a row of digits above the letters. The option is in the on-screen keyboard settings; the band grows a little taller. • The app now resizes above the keyboard. The send button and the last lines of text used to stay buried under the keys, out of reach until the keyboard was closed. • Spreading two fingers to reach the braille keyboard works at last. The distance it asked for was wider than the screen, so the gesture fell through to the action below it and started a dictation. • That dictation can be stopped again. A two-finger double tap always stops it, where it could start media playback while the microphone stayed open, and locking the phone was the only way out. • No more letter announced twice and followed by "end of field" on every keystroke, with Gboard as with a physical keyboard. And no more double echo in braille in web fields, such as Gmail's compose. Braille keyboard • The dots and the swipes follow how you hold the phone. With the layout setting on "adapt to how you hold the device", the accelerometer decides, live, and a change of posture is announced. An explicit choice, flat or screen away, always wins. Before, anything that was not explicit was treated as flat on a table: anyone holding their phone had the dots and the swipes mirrored, a finger to the left making a space. • Turning the phone half a turn in landscape, charging port on the other side, now turns the dot grid and the swipes with it. Recent apps • The menu no longer goes through a submenu: activating a row opens the app. Pinning and the system app details are now actions on the row, offered by the actions rotor, or by a double tap and hold that opens them as a list. Pinning leaves the menu open; the app details screen closes it first. • The menu remembers thirty apps instead of twelve. • "Previous app" and "next app" are more predictable: the walk is always anchored on the app you are in, so the first gesture always lands on the one used before it. Only apps that can actually be launched enter the list, which keeps permission dialogs and system components out of it. Speech • The time, the battery, the window title, the phonetic spelling and the other announcements you ask for with a gesture are no longer dropped while media is playing, a microphone is open or a call is up. The gesture worked and so did the action: only the sentence was thrown away. Settings • Sound settings no longer crash the reader on a device with no vibrator. The vibration-strength slider added in the previous version depended on a switch those devices do not show. • On a dialer keypad, resting a finger on a key to hear it no longer triggers a long press: "1" no longer calls voicemail and "0" no longer becomes a plus. Typing on lift works, and a deliberate long press is still available by double-tapping and holding. • On a physical keyboard, the arrow keys move from item to item, like a swipe, whatever the rotor is set to. Scripts • A script can declare its own options, with their type and default. They are set in "My scripts", and also straight from the Talk Forward menu of the app the script runs in, so they are not a trip through Settings mid-task. • Scripts get a clock, tf.now(), so they can rate-limit themselves in a chatty app. Help and translations • The help finally describes the gestures that are actually in force. Sending from the braille keyboard is three fingers up, not two; hold to dictate is off by default; the scrub needs a single back-and-forth; the rotor has its groups; the two-finger swipes list their real defaults; and a section describes the letter keyboard. • The braille keyboard gestures page describes the keyboard you actually type on, and the gestures you customize there finally apply. • Every string is translated into the app's 75 languages: about twenty thousand texts were still showing in English. In Simplified Chinese, the dictation engine list could show the wrong engine.

Version 26.9.2

Released September 14, 2026Android Talk Forward 26.9.2 Recent apps • Talk Forward now keeps its own list of recently used apps, with three assignable actions built on it. "Previous app" and "next app" walk the list Alt-Tab style: swipes chain together, and the app you stop on becomes the most recent. "Recent apps" opens the list as an ordinary Talk Forward menu — explorable by touch, navigable by swipe, closed by Back, spoken in the reader's own voice. Each app there offers open, pin or unpin, and its system app-info screen. A pinned app heads the list whether or not it has been used lately. No gesture is bound by default: the three actions are waiting at the bottom of the action list. • The list is observed, never queried: an app enters it after staying in the foreground for about a second, which keeps out transient activities, share sheets and permission dialogs. Home screens, the system UI, the active keyboard and Talk Forward itself never enter it. It survives a restart. Actions inside an app • A gesture can now do something in an app rather than merely open it: call a particular contact, open a particular conversation, start a route. Pick "Action in an app" from a gesture's action list: Talk Forward lists the apps that can build one, and it is the app's own chooser that opens to let you pick the contact or the conversation. Nothing about its interface is guessed, so nothing breaks when it is updated. Coverage depends on the apps: on a Pixel, Contacts, WhatsApp, Maps, Gmail, Docs and Settings all offer one. • Fixed along the way: in the per-app gesture editor, choosing "Launch application" stored a gesture that did nothing. Vibration • A slider sets how strong every vibration of the reader is, from a quarter to double its normal strength. It sits in Sound and vibration, under the vibration switch. On a device whose vibrator cannot vary its strength, the patterns play as authored rather than pretending. • On-screen braille typing finally vibrates, as the letter keyboard already did. Both keyboards share one setting, renamed "Typing vibration". Braille cells get the lighter tick: a cell lands far more often than a key. Physical keyboard • Ctrl, Alt and the Windows key can now serve as the Talk Forward modifier, alongside CapsLock, Insert and Scroll Lock. Left and right stay separate choices, so binding one leaves the other doing its usual job. Note that pressing Ctrl interrupts speech, a behaviour independent of keymaps, and that Android reserves some Windows-key combinations for itself. Explore by touch • The finger and the swipe select the same thing again. On a rich row — a message in the Gmail widget, a notification — exploring by touch selected a fragment of the row, its sender or its subject, where a swipe selected the whole row. The next swipe then carried on from the fragment. • The finger no longer falls into holes. A widget's margins and the few pixels between two rows belong to the container: crossing them cut the message being read to announce "Gmail", over and over. Those areas are now treated as a hole in the item being read, not as a new item. • An announcement is no longer cut off by the hover that follows it on the same item. In the notification shade a notification would start being read and be interrupted thirty milliseconds later by another hover on that same row: nothing was ever heard. • On a screen-filling backdrop — the home screen page under the icons — the description is no longer repeated every time the finger lands back on empty space. It is spoken on arrival, then when it changes, which is when you change page. In between, a regular tick tells you the finger is on empty space. Braille • Correcting a word mid-typing no longer strands the start of the word in the field. Deleting a cell and retyping produced "dedernières" instead of "dernières", because the deletion committed what was being composed. Ringer • The rotor now tells the truth about silent mode: on Android an app can only reach it by turning Do Not Disturb on. The announcement says so instead of implying plain silence.

Version 26.9.1

Released September 12, 2026Android Talk Forward 26.9.1 Visual agent • The agent now answers follow-up questions. When it has finished speaking, the microphone stays open for five seconds: carry on without touching anything. "Tell me more about that button", "and just below it?", "press it" — your previous question and its answer go with the new one, so it knows what you are referring to. Say nothing and the window closes silently: no beep, no prompt to sit through when the answer was all you wanted. The five seconds are counted from the moment the microphone opens, not from the start of the answer, so a long description never eats into your time to speak. • The agent also performs the Talk Forward actions you name. "Next item", "home", "turn screen recognition on" — everything you can assign to a gesture is recognised by its name, in your language, without going through our servers. The matching is deliberately strict so that it never confiscates your questions: "what is the next button?" is still a question for the agent. Sound • The screen reader volume can follow the media volume, and the other way round. The option is in Sound and vibration, under the volume slider. With it on, moving either one moves the other, whether you use the volume keys, the slider in the settings, or Android's own audio settings. The two do not have the same number of steps, so it is the relative level that is matched. On-screen keyboards • A two-finger scrub closes the keyboards. The gesture that means "back" elsewhere now works on the letter keyboard band as well, and during braille input, where it closes for good: no other keyboard takes its place. • Letter keyboard: what you touch and what your swipes act on can no longer drift apart. After exploring an element of the app above the keyboard, a swipe right still announced the next key. The keyboard and the screen reader now share a single cursor: exploring a key brings it onto the keyboard, exploring or reaching an element of the app brings it onto the app, and a swipe acts on whichever of the two you are actually on — whether the gesture is made on the keys or above them. A new setting, in the on-screen keyboard preferences, decides where the cursor starts when the keyboard opens. • Braille input: in number mode, a cell is no longer announced twice under two names. The number prefix followed by dots 1-6 said both "1" and "a circumflex". The keyboard's announcement now asks the edit buffer what the cell will actually write in its context, instead of translating the cell on its own. Capitals benefit too: the capital prefix followed by a letter announces the capital. • Braille input: no more "end of field" after every cell you type. Each insertion makes the field report that the cursor moved, and the screen reader announced that move on top of the keyboard's own echo. Cursor moves you ask for are of course still announced. • Braille input: browsing autofill suggestions no longer sticks. Once opened on an address or password field, it survived the keyboard closing, and every later session started inside it — swipes up and down cycled suggestions instead of moving the cursor, with nothing to say why. It now closes with the keyboard, and typing a cell leaves it. Calls • The caller's name can be repeated for as long as the phone rings. A single announcement gives you a single chance: the ringtone starts at the same moment, or you are not next to the phone, and the only way left to find out who is calling is to pick it up. You choose what is announced — the name, the number, or both — and how often it comes back. • Answering and hanging up now work for calls from messaging apps, with the two-finger double tap. A Telegram or WhatsApp call is not a phone call as far as Android is concerned: the system saw it ring and then lost sight of it, and the instruction to answer was lost with it. Talk Forward now presses the button on the app's own call notification, exactly as you would, and only turns to the system afterwards. Touch exploration • No more items that vibrate and play their sound without saying anything. It happened a great deal in the notification shade, but also on a widget or in a web page, and never when swiping. Two separate causes: an item that already held the focus was not announced at all when the finger arrived on it, and an expanded group — a notification you have opened up, for instance — swallowed everything inside it, so the finger could travel over several texts without the cursor ever moving. Inside an item that has already been announced, exploration now goes down to what is really under your finger. Play and pause • The two-finger double tap finally works everywhere. It did not pause Prime Video and did not answer on WhatsApp, while a button on a wired remote did both. Talk Forward now drives whatever is playing directly, the way the buttons on the player's notification do, and falls back to a real media key when there is nothing to drive. For calls, the "answer" key of hands-free kits is sent first; if the call has not moved a moment later, the precise routes take over. Rotor • The rotor says when it has nothing to offer. Turning it with no reading control available made no sound at all, which is indistinguishable from a gesture that was not recognised, from a broken feature, and from a screen reader gone mute. Dictation • The field's placeholder no longer ends up in front of your text. Dictating into an empty message field inserted "Message hello, what time…". Apps that show that prompt text without declaring it as such are now recognised, and the field is treated as empty. Translations • The text added by this version is translated into the 28 languages Talk Forward already covers.

The fact that this was originally released all the way back in June and no one has brought it up till now (which you can check out it's changelog here , is actually quite shocking!

Download can be found via the main Talk Forward site by navigating to the "Desktop companion" heading (which is at level 3).