Kyra vs Saudi — does the dialect hold?
Your two voices first, the built-ins after the gap as a second opinion. Six words that expose the ق. Nothing here is in the app yet.
What to listen for
Palestinian drops ق to a glottal stop: ’ahwe, soo’, wa’it.
Saudi keeps it as a hard /q/: qahwa, sooq, waqit. If the two
columns sound the same on those three rows, the dialect parameter is not doing
what it claims and none of these are worth shipping. The last row has no ق at
all and should sound alike either way — that is the control.
| Word | Kyra Palestinian · yours | Saudi Saudi · yours | Layan pls built-in | Amjad pls built-in | Hiba ksa built-in | |
|---|---|---|---|---|---|---|
قَهْوَة want ’ahwe · ق initial |
||||||
سُوق want soo’ · ق final |
||||||
وَقِت want wa’it · ق medial |
||||||
حَدِيقَة want hadii’a · between vowels |
||||||
قَلْقَان want ’al’aan · ق twice |
||||||
مَبْسُوط control · no ق |
Why this provider is different. Every Levantine voice in the app so far has
been Levantine by inheritance. Hamsa takes a
Nothing is committed. These 24 clips are a comparison, not a voice. Rendering one voice into the app is 363 more calls; say which and it goes in.
One fault already seen. Fahd's قَهْوَة came back as 0.23s of digital silence with a plausible file size, and the identical request a moment later was fine. The renderer now retries on silence — without that the word would have been dropped to the device voice permanently, which is how two of Nour's clips were lost.
dialect parameter whose
enum lists pls, syr, jor and leb
as four separate entries rather than one bucket — a model that distinguishes
Palestinian from its neighbours instead of averaging them.
Nothing is committed. These 24 clips are a comparison, not a voice. Rendering one voice into the app is 363 more calls; say which and it goes in.
One fault already seen. Fahd's قَهْوَة came back as 0.23s of digital silence with a plausible file size, and the identical request a moment later was fine. The renderer now retries on silence — without that the word would have been dropped to the device voice permanently, which is how two of Nour's clips were lost.