Title: Incorrect pronunciation of frequently used Japanese words

Bug Description:

Frequently used Japanese words such as "こんにちは" (Konnichiwa - hello), "はい" (Hai - yes), and "いいえ" (Iie - no) are not pronounced correctly.

Environment:

• StackFlow 1.8
• llm-melotts/stable,now 1.7 arm64
• llm-model-melotts-ja-jp/stable,now 0.5 arm64
• apt upgrade was performed after the StackFlow 1.8 release.

Steps to Reproduce:

1. Install software that uses M5Module-LLM on an M5StackBasic device. Connect the output of the llm-llm module to the llm-melotts module.
2. In the ApiMelottsSetupConfig_t structure, set config.model to melotts-ja-jp.
3. Provide input to the LLM that results in output phrases starting with Japanese words like "こんにちは" (Konnichiwa - hello), "はい" (Hai - yes), or "いいえ" (Iie - no).

Comments on Reproducibility:

Currently, this issue seems to occur primarily when words written entirely in hiragana (Japanese phonetic characters) appear at the beginning of a sentence. However, it's important to note that the problem doesn't manifest in all instances of hiragana-only words.

Observed Behavior:

A clear example of this mispronunciation is when the word "こんにちは" (Konnichiwa - hello) is spoken; it sounds like "かんこば" (kankoba). A video demonstrating this behavior can be found at the following link: https://x.com/nyasu3w2022/status/1922263985421774881

Code Causing the Issue (M5StackBasic side):

The code running on the M5StackBasic that exhibits this issue can be found here: https://gist.github.com/nyasu3w/2472c4bd7b397c3efff75a146465be63

p.s.:

• It seems that numbers are also not pronounced in Japanese.
• This issue was created with the support transration of Manus(AI).