Voice - Announce artist & track name

[dorotik]

Since there is the "Announce Battery Level" feature, it shouldn't be too hard to add the "Announce artist & track name" feature. It would be a great help for remembering song names.

I'll wait till somebody from the Rockbox dev. team responds whether this is possible or not. If so, I can help implementing it.

[bluebrother]

Since there is the "Announce Battery Level" feature, it shouldn't be too hard to add the "Announce artist & track name" feature. It would be a great help for remembering song names.

It is hard and currently not feasible.

Rockbox uses prerendered voice clips. For battery level this is easy since there is a limited number of voice strings which are known in advance. For track metadata like artist and track name this is not true. For the file browser to be able to voice filenames there are two options: (1) spell filenames, in which case only the alphabeth is needed. However, this adds another problem: non-ASCII characters, which are currently not spelled -- givent he number of characters in use in the world the list of characters would become huge. And (2) use prerendered "talk" clips. This has the drawback that the user needs to remember to (re)generate them when changing the files on the player.

So to implement voicing metadata there are two options: (1) roll a new talkfile like system for metadata. This means (a) yet another system for voicing and (b) has the same drawbacks as talkfiles have right now. Or (2) implement a TTS into Rockbox. While people do agree this to be a good thing it's really hard -- the limited resources on the players supported by Rockbox really makes this problematic. Additionally, this means people would be bound to the voice the integrated TTS provides, and that voice is unlikely to be comparable to desktop PC voices. Some years ago there was a GSoC project trying to integrate a TTS into Rockbox. Unfortunately it failed.

[dorotik]

Since there is the "Announce Battery Level" feature, it shouldn't be too hard to add the "Announce artist & track name" feature. It would be a great help for remembering song names.

It is hard and currently not feasible.

Rockbox uses prerendered voice clips. For battery level this is easy since there is a limited number of voice strings which are known in advance. For track metadata like artist and track name this is not true. For the file browser to be able to voice filenames there are two options: (1) spell filenames, in which case only the alphabeth is needed. However, this adds another problem: non-ASCII characters, which are currently not spelled -- givent he number of characters in use in the world the list of characters would become huge. And (2) use prerendered "talk" clips. This has the drawback that the user needs to remember to (re)generate them when changing the files on the player.

So to implement voicing metadata there are two options: (1) roll a new talkfile like system for metadata. This means (a) yet another system for voicing and (b) has the same drawbacks as talkfiles have right now. Or (2) implement a TTS into Rockbox. While people do agree this to be a good thing it's really hard -- the limited resources on the players supported by Rockbox really makes this problematic. Additionally, this means people would be bound to the voice the integrated TTS provides, and that voice is unlikely to be comparable to desktop PC voices. Some years ago there was a GSoC project trying to integrate a TTS into Rockbox. Unfortunately it failed.

Sorry, maybe I failed to explain this properly. I'll give an example:

I'm using the Rockbox utility to generate .talk clips using a TTS from NeoSpeech. On my target I enable the "Use File .talk Clips" feature. From now on, Rockbox will play a .talk clip for voiced tracks whenever I'm scrolling through playlist or files. It plays it without interrupting the playback, so technically it could be possible that Rockbox would play the .talk clip automatically during the beginning of each track. If some track does not have a .talk clip, nothing gets played - no problem.

Basically I should have named this feature "Announce File", not "Announce artist & track name", since it's the file name which gets voiced.

[gevaerts]

I would imagine that to be doable. You'd need to call talk_file() with the right arguments whenever the track changes, which you should be able to detect by registering a callback for the PLAYBACK_EVENT_TRACK_CHANGE event. Add the necessary settings and you're done.

[bluebrother]

Basically I should have named this feature "Announce File", not "Announce artist & track name", since it's the file name which gets voiced.

That's something pretty different and, as has been said, much more doable. However, I'm not sure about the actual use for it -- not all people have their files named properly, and even if it's not easy to say what "properly" is. Plus, when listening to music I want to listen to music, not the music getting distorted by announcements

If someone provides a patch for it people will consider it, but I don't think you'll find much interest in this among developers.