Researching about speech-to-text, text-to-speech, and LLAMA API |
• Learned that the original plan to use google tts and stt had a difficult registration, and potentially hit the limit and paid immediately (300 dollars, 5 million). Google tts and stt have a free limit of 4 million characters per month, but must enter credit and NIK for payment if they exceed the limit. |
• Learned that LLAMA API has 2 options, build your own model or use a third party, such as Replicate, together.ai, and fireworks.ai. |
|
• Have explored alternative free tts and sst besides Google. |
|
tts : eSpeak NG (robotic language, not in API form), Coqui (Must use python with pyTorch/Tensorflow library). |
|
sst: Vosk (lightweight, fits the project architecture diagram), Whisper by OpenAI, Bookbot-kids speech recognizer. |
|