Udio Specialists alkalmazható
1 eredmény megjelenítése
Szponzorált szabadúszók
-
Dolgozzon vele: asmaaw66
asmaaw66 asmaaw66
$20 USD / óra
Smart Systems Engineer | MATLAB & DSP SpecialistPalestinian TerritoryStrategy for Answering the Client's Questions 1. Handling Silence/Noise: The Approach: Propose using a Short-Time Energy (STE) and Zero-Crossing Rate (ZCR) thresholding method for the VAD. The Logic: You will process audio in chunks (e.g., 5-second blocks). If the energy in a block falls below a calculated floor...Strategy for Answering the Client's Questions 1. Handling Silence/Noise: The Approach: Propose using a Short-Time Energy (STE) and Zero-Crossing Rate (ZCR) thresholding method for the VAD. The Logic: You will process audio in chunks (e.g., 5-second blocks). If the energy in a block falls below a calculated floor (calculated relative to the noise floor of that specific file), the script skips that block. This ensures the "Yes/No" decision is based strictly on speech characteristics, not dead air. 2. Features for TSM Detection: The Approach: Focus on Phase Variance or Formant Trajectory continuity. The Logic: Time-Scale Modification (TSM), especially if done poorly, often introduces phase discontinuities or unnatural smoothing in formant trajectories. By tracking the variance of these features across a long duration, you can detect the "periodic" artifacts left by the stretching algorithm. Draft Proposal (Copy & Paste) Subject: Expert MATLAB DSP Developer for Blind TSM Detection Hi there, I am a Smart Systems and Devices Engineering student with a deep specialization in Digital Signal Processing (DSP) and system stability analysis. I have extensive experience in MATLAB and have recently worked on signal-based projects involving sensor logic and hardware-level signal conditioning. I am very interested in your audio forensics project. My approach to your project: 1. Handling Silence and Noise: To handle the 90-minute files without skewing results, I will implement a VAD pre-filter using a dynamic energy-based thresholding approach. By calculating the signal-to-noise floor for each file dynamically, the script will ignore blocks below the threshold. This ensures that the detection features are extracted only from valid speech segments, preventing silence from diluting the statistical variance required for classification. 2. Feature Extraction for Blind TSM Detection: For the 90-minute timeline, I propose monitoring the Short-Time Phase Variance and Formant Trajectory smoothness. TSM processes, especially those involving granular synthesis or synchronous overlap-add (SOLA), often leave distinct periodic signatures in the phase spectrum. By analyzing the deviation of these features over time against a calibrated threshold, we can accurately flag synthetic scaling artifacts. Experience: My academic background includes advanced training in Z-transforms, convolution, and system stability analysis, which are essential for identifying signal artifacts. I am comfortable working with dsp.AudioFileReader to ensure memory-efficient, block-based processing, even for long-form audio. I am confident I can provide a robust, commented script that meets your requirements. I am available to start immediately and would love to discuss the specific artifacts you are most concerned about. Best regards, Asmaa Alnabhan kevesebb
