Voice model for neural network

Hello! I need to improve the quality of an audio recording for clear understanding of the words. Here are the task details: The conversation of a person is poorly audible due to strong background noise from a car engine. The engine sound is monotonous but drowns out the useful signal (speech). Additionally, the voice sounds muffled as the conversation was conducted through a speakerphone. The duration of the recording is approximately 15-20 minutes. Maximize the suppression of the low-frequency rumble of the engine, raise the intelligibility frequencies of speech (range 1–4 kHz), and make the words clear for transcription (minutes). Special requests: It is important not to "squash" the sound with AI filters so that the voices do not turn into robotic metallic artifacts. If it is impossible to completely remove the rumble without harming the speech, it is better to leave some noise but preserve the articulation of the words and ensure that both speakers are understandable.

Transcribe the audio file. Convert it into printed text while preserving the exact meaning. The audio has noise. Duration - about 1 hour.

Audio 30 minutes with noise. Transcribe the conversation of each participant into text, excluding sounds from the television and the street.

Hello everyone! I'm looking for a BeatMaker with a portfolio in the Hip Hop / Trap genre. Please send links to your work. Thank you.

I have an AI voice agent that speaks Spanish using predefined scripts. The goal is to make its speech sound as natural and human as possible. Your task is to have conversations with the AI and identify any issues that make it sound unnatural, including logic, grammar, syntax, pronunciation, filler words, pauses, intonation, or any other nuances. For each issue, you'll provide the exact phrase spoken by the bot, classify the issue (e.g. logic, syntax, pronunciation, pauses, intonation), and suggest a natural Spanish alternative. We then update the script and repeat the process until the conversation reaches production quality. The final version is reviewed by another native Spanish speaker before deployment. There are several call scenarios available, and more will be added over time. You'll also have access to a web interface where you can edit prompts, launch calls yourself, and test improvements directly. Compensation is flexible and can be hourly or task-based, depending on the assignment. In addition, performance bonuses are available if the final version performs well in real customer calls with a low AI detection rate.