What's New

• Low precision (fp16) execution for Voice Isolation Models: set enable_low_precision_execution: true to reduce CPU usage by roughly 38–50% on avx512 and armv8a hardware. Hardware support is detected automatically at runtime, so callers don't need to check for it themselves on unsupported hardware or models, the flag is simply a no-op and processing continues at native precision. Defaults to false.
• Accent Detection API: requires approximately 1 second of audio to produce a result.
• Gender Detection API: requires approximately 4 seconds of audio to produce a result.
• Compatible with VIVA C/C++ SDK 9.21.0 and above.

⚠ Breaking Change

• NcSessionConfig gained a new field, enable_low_precision_execution: bool. This struct
  has no Default implementation, so any existing code that constructs it as a full struct
  literal will fail to compile until this field is added. Set it to true to enable CPU optimization on avx512 and armv8a hardware.