Voice Nav Bugs, Scenic 3
Quote from RichS on 21.11.2021, 20:08Hi - been enjoying Scenic a lot - tried the new version the other day. I ran into a few issues with voice instructions. The text-to-speech seems to get confused with typical navigation routing words. For instance, for a left turn onto MA Rt-225 the voice pronounces it like this: "Now turn left onto Mah AreTee Negative Two Twenty Five" where MA (Massachusetts) comes out as 'Mah", Rt becomes "AreTee" and the hyphen becomes "Negative". A turn onto U.S. Rt. 3 might come out as "Us AreTee 3". I'm on an iPhone 13 and IOS 15.1, and listening through a Sena 30K via Bluetooth in my helmet. I can imagine some sort of dictionary-driven word and punctuation replacement between the navigation instructions and the text to speech function to replace something like Rt. with Route, to eliminate the hyphens, etc. Scenic 2 had this problem a bit but not as much as Scenic 3. I didn't report it then because you said you were swapping out the whole nav module so I figured I'd just wait and see, but the new version does it too, and worse.
I also am running into the problem where some voice instructions do not come out at all - it might tell me 1000 ft. before a turn, "In 1000 feet turn left" but then omit the instruction when I approach the road "Now turn left". I see some others have reported this. This happened on only a few of the turns but I couldn't spot a pattern.
I am also running into the problem where the first bit of voice instruction is cut off, as the Bluetooth to the Sena 30K shuts off between instructions and does not instantly start up when Scenic sends sound again, so the first word or two might get chopped. This did not happen to me using Scenic 2 with the same phone/helmet combo.
Thanks again for such an excellent app! I use it all the time - it really has been terrific.
Hi - been enjoying Scenic a lot - tried the new version the other day. I ran into a few issues with voice instructions. The text-to-speech seems to get confused with typical navigation routing words. For instance, for a left turn onto MA Rt-225 the voice pronounces it like this: "Now turn left onto Mah AreTee Negative Two Twenty Five" where MA (Massachusetts) comes out as 'Mah", Rt becomes "AreTee" and the hyphen becomes "Negative". A turn onto U.S. Rt. 3 might come out as "Us AreTee 3". I'm on an iPhone 13 and IOS 15.1, and listening through a Sena 30K via Bluetooth in my helmet. I can imagine some sort of dictionary-driven word and punctuation replacement between the navigation instructions and the text to speech function to replace something like Rt. with Route, to eliminate the hyphens, etc. Scenic 2 had this problem a bit but not as much as Scenic 3. I didn't report it then because you said you were swapping out the whole nav module so I figured I'd just wait and see, but the new version does it too, and worse.
I also am running into the problem where some voice instructions do not come out at all - it might tell me 1000 ft. before a turn, "In 1000 feet turn left" but then omit the instruction when I approach the road "Now turn left". I see some others have reported this. This happened on only a few of the turns but I couldn't spot a pattern.
I am also running into the problem where the first bit of voice instruction is cut off, as the Bluetooth to the Sena 30K shuts off between instructions and does not instantly start up when Scenic sends sound again, so the first word or two might get chopped. This did not happen to me using Scenic 2 with the same phone/helmet combo.
Thanks again for such an excellent app! I use it all the time - it really has been terrific.
Quote from Guido on 22.11.2021, 09:48Hi Rich,
Thanks for letting me know. I'll have to report this first issue to the map provider. They use the same synthetic voice algorithm (comes with iOS) as the old framework, but from the looks of it they are using abbreviations without dots in between... thus the synthetic voice is interpreting and attempting to pronounce as words and not separate letters. For now, the only way around this, is to select a pre-recorded voice that doesn't speak street names. I know this is not really a solution 🙂
I think the second problems happens when there's a via point very close to or on a junction. This will be resolved with the upcoming update.
The 3rd problem... do you have 'beep before instruction' turned on? (Settings > Sound & Haptics)? If not, that might help. If you already do, let me know. Do you hear part of that sound, or even with this turned on is the first bit of the spoken instruction cut off? The new audio engine has an option to add a 'silent sound' before the spoken voice instruction. I just did not include this as a config option as I thought it wouldn't be necessary with the 'beep before instrcution'.
Hi Rich,
Thanks for letting me know. I'll have to report this first issue to the map provider. They use the same synthetic voice algorithm (comes with iOS) as the old framework, but from the looks of it they are using abbreviations without dots in between... thus the synthetic voice is interpreting and attempting to pronounce as words and not separate letters. For now, the only way around this, is to select a pre-recorded voice that doesn't speak street names. I know this is not really a solution 🙂
I think the second problems happens when there's a via point very close to or on a junction. This will be resolved with the upcoming update.
The 3rd problem... do you have 'beep before instruction' turned on? (Settings > Sound & Haptics)? If not, that might help. If you already do, let me know. Do you hear part of that sound, or even with this turned on is the first bit of the spoken instruction cut off? The new audio engine has an option to add a 'silent sound' before the spoken voice instruction. I just did not include this as a config option as I thought it wouldn't be necessary with the 'beep before instrcution'.
Quote from RichS on 22.11.2021, 10:45Hey Guido,
Thanks for the quick answer! I have not tried it with "beep before instruction" turned on - I will try that next time I'm on my bike and let you know.
On the first problem - it seems like the text-to-speech isn't expanding abbreviations into words, and is not interpreting punctuation intelligently for navigation, but rather for more general uses. Whether "Rt" or "Rt." it shouldn't pronounce it "aretee" but rather "route". And hyphens should never be pronounced "negative" when doing navigation. But the opposite is true of "US 3" - US should never be pronounced "us", but rather U.S. when navigating. Likewise, "MA" is "Massachusetts" and never "mah". If the text-to-speech is integral with the maps, then I'm really surprised, as this bug would be happening everywhere this map provider's APIs are in use, not just Scenic. (As you can probably guess, I have a career-long history in the software industry). Isn't stuff like this usually handled by domain-specific translation tables in the text-to-speech pipeline? Medical equipment has the same problem - medical-speak is hard to pronounce but follows specific rules that can go between the word generation and the speech generation. Anyway - hope this gets corrected, though some of the mispronunciations are pretty funny. 🙂
Thanks again!
Hey Guido,
Thanks for the quick answer! I have not tried it with "beep before instruction" turned on - I will try that next time I'm on my bike and let you know.
On the first problem - it seems like the text-to-speech isn't expanding abbreviations into words, and is not interpreting punctuation intelligently for navigation, but rather for more general uses. Whether "Rt" or "Rt." it shouldn't pronounce it "aretee" but rather "route". And hyphens should never be pronounced "negative" when doing navigation. But the opposite is true of "US 3" - US should never be pronounced "us", but rather U.S. when navigating. Likewise, "MA" is "Massachusetts" and never "mah". If the text-to-speech is integral with the maps, then I'm really surprised, as this bug would be happening everywhere this map provider's APIs are in use, not just Scenic. (As you can probably guess, I have a career-long history in the software industry). Isn't stuff like this usually handled by domain-specific translation tables in the text-to-speech pipeline? Medical equipment has the same problem - medical-speak is hard to pronounce but follows specific rules that can go between the word generation and the speech generation. Anyway - hope this gets corrected, though some of the mispronunciations are pretty funny. 🙂
Thanks again!
Quote from Guido on 22.11.2021, 11:03I understand what you're saying. What I was trying to say is that this is out of my direct control. The framework determines the text that needs to be spoken, and the audio engine (which synthesizes this text into speech) is also provided by the framework. So, unless there's a configuration option that I missed, this is a bug in the framework. I have reported this to the provider now, and will have to wait what they come back with.
I understand what you're saying. What I was trying to say is that this is out of my direct control. The framework determines the text that needs to be spoken, and the audio engine (which synthesizes this text into speech) is also provided by the framework. So, unless there's a configuration option that I missed, this is a bug in the framework. I have reported this to the provider now, and will have to wait what they come back with.
Quote from Guido on 26.11.2021, 08:55Hi Rich,
So, I did some more research on this and talked to the framework provider. Long story short... it's probably due to the voice you have selected in iOS Settings > Accessibility > Spoken Content > Voices.
Whenever you select a synthetic voice for a certain language in Scenic, (like the 'English (US) announced street names' voice), then the framework will use the voice you selected for that language in the Spoken Content voices. I've tried this and selected 'Samantha' for English US. Samantha pronounces St as Street and Rd as Road. I haven't been able to find a route where it tried to pronounce "MA Rt-225".
Could you give it a try by selecting the 'Samantha' voice like described? If this doesn't help / still pronounces wrong, could you perhaps share the link to the route where you see/hear this happening (select route from your route list and then use 'share' button)? Then I can simulate the exact situation here.
I hope that this is resolved by selecting another voice though.
Cheers,
Guido
Hi Rich,
So, I did some more research on this and talked to the framework provider. Long story short... it's probably due to the voice you have selected in iOS Settings > Accessibility > Spoken Content > Voices.
Whenever you select a synthetic voice for a certain language in Scenic, (like the 'English (US) announced street names' voice), then the framework will use the voice you selected for that language in the Spoken Content voices. I've tried this and selected 'Samantha' for English US. Samantha pronounces St as Street and Rd as Road. I haven't been able to find a route where it tried to pronounce "MA Rt-225".
Could you give it a try by selecting the 'Samantha' voice like described? If this doesn't help / still pronounces wrong, could you perhaps share the link to the route where you see/hear this happening (select route from your route list and then use 'share' button)? Then I can simulate the exact situation here.
I hope that this is resolved by selecting another voice though.
Cheers,
Guido
