Sophisticated with large gaps
The software doesn’t tell you why you don’t pronounce correctly. You are shown statistical “closeness” percentages which you can find yourself altering pronunciation until you arrive at the correct accent, but it’s a very poor feedback loop. Because there is no “classification” to tell you what mistake you’re making, you don’t know.
This is difficult when your accent is close, but not perfect.
Additionally, the software is highly dependent on good audio input. It even regularly fails to confidently understand what you’re saying when using voice isolation from your iPhone’s mic, let alone bluetooth input, which may be responsible for poor audio input due to the way the developer receives that sound.
Finally, when attempting to compare the reference phonemes to your own, the spliced audio is cut incorrectly almost always, so you hear bursts of incomplete incomprehensible audio instead of a fully intelligible phoneme.
All of these are
major failure points. Together, they erode user experience in a way that makes the app a great technical demonstration, but not a practical piece of software for improving your Parisian accent.
The voice models used to get the app this far are clearly sophisticated, but every single part of the app has to be well done, not just the voice models analysis or look and feel of the colors, typesetting, and artwork used.
For example, when you hear a person of Indian heritage speak American English, as a native American English speaker, you know precisely what makes their characteristic accent an improper, say, West Coast American English accent. And you could point it out.
This app does not do that. It would also presumably complicate the architecture of their software behind the scenes. But, you need this type of information. Not abstract percentages.
I want to see explicit correction, not percentage confidence that machine learning engineers so easily fallback on even though it’s meaningless to users.
I want better audio processing. You can currently see how badly it hears you by shrinking some words when it has low confidence in what you said.
I want to hear complete, intelligible phonemes, not bursts of noise when I tap.
And finally, not mentioned here above, I want better graphs to see my progress. Right now, you’re given a score over time, but you can’t tap on it and see what’s going on or why. The best you get is a dumped list of these abstract percentages in a generic list with all of the phrases you’ve pronounced.
There’s a lot of promise in this app. To my current knowledge, it’s the best thing on the market. But, it’s also not usable. Maybe the technology isn’t there yet, but I doubt it. This seems to me to be a matter of implementation
improvement.
This is difficult when your accent is close, but not perfect.
Additionally, the software is highly dependent on good audio input. It even regularly fails to confidently understand what you’re saying when using voice isolation from your iPhone’s mic, let alone bluetooth input, which may be responsible for poor audio input due to the way the developer receives that sound.
Finally, when attempting to compare the reference phonemes to your own, the spliced audio is cut incorrectly almost always, so you hear bursts of incomplete incomprehensible audio instead of a fully intelligible phoneme.
All of these are
major failure points. Together, they erode user experience in a way that makes the app a great technical demonstration, but not a practical piece of software for improving your Parisian accent.
The voice models used to get the app this far are clearly sophisticated, but every single part of the app has to be well done, not just the voice models analysis or look and feel of the colors, typesetting, and artwork used.
For example, when you hear a person of Indian heritage speak American English, as a native American English speaker, you know precisely what makes their characteristic accent an improper, say, West Coast American English accent. And you could point it out.
This app does not do that. It would also presumably complicate the architecture of their software behind the scenes. But, you need this type of information. Not abstract percentages.
I want to see explicit correction, not percentage confidence that machine learning engineers so easily fallback on even though it’s meaningless to users.
I want better audio processing. You can currently see how badly it hears you by shrinking some words when it has low confidence in what you said.
I want to hear complete, intelligible phonemes, not bursts of noise when I tap.
And finally, not mentioned here above, I want better graphs to see my progress. Right now, you’re given a score over time, but you can’t tap on it and see what’s going on or why. The best you get is a dumped list of these abstract percentages in a generic list with all of the phrases you’ve pronounced.
There’s a lot of promise in this app. To my current knowledge, it’s the best thing on the market. But, it’s also not usable. Maybe the technology isn’t there yet, but I doubt it. This seems to me to be a matter of implementation
improvement.
Show more
