NVIDIA Riva and NeMo Speech Quiz

Answers and explanations appear as you take the quiz.

1. What is the primary distinction between NeMo and Riva?
   • NeMo is closed-source while Riva is open-source
   • NeMo is for training and fine-tuning models, while Riva is for optimized deployment and serving
   • NeMo only handles text-to-speech and Riva only handles recognition
   • They are two names for the exact same product
2. Which NVIDIA components does Riva documentation identify for optimized model execution and inference serving?
   • TensorRT for optimizations and Triton Inference Server for serving
   • OpenGL for language modeling and a SQL database for waveform decoding
   • Only CUDA kernels with no inference server
   • A browser rendering engine and an image-editing service
3. In a typical Riva TTS pipeline, what is the role of a vocoder like HiFi-GAN?
   • It converts speech audio into text
   • It detects which speaker is talking
   • It translates text between languages
   • It converts acoustic features (like mel-spectrograms) into an audio waveform
4. What does 'streaming' ASR in Riva enable?
   • Transcription only after the full recording is uploaded
   • Video streaming without any audio processing
   • Partial transcripts returned as audio arrives, with very low latency
   • Downloading models faster over the network
5. Which is an example of customization NeMo enables for speech models?
   • Rewriting the GPU firmware
   • Fine-tuning an ASR model on domain-specific vocabulary such as medical terms
   • Encrypting audio at the network layer
   • Compressing video files
6. Which service interface does Riva expose for speech requests in the documented deployment?
   • A local spreadsheet macro interface only
   • A graphics-driver rendering API only
   • A text tokenizer with no service endpoint
   • gRPC-based microservices
7. Before exporting a fine-tuned NeMo model to Riva, what compatibility check is necessary?
   • Assume every NeMo checkpoint is directly interchangeable with a Riva model
   • Check only the checkpoint's file name
   • Confirm that the architecture, export path, and software versions are supported
   • Assume an ASR model can be deployed as a vocoder without conversion
8. How should a team determine whether Riva streaming ASR meets its latency target under many simultaneous calls?
   • Measure the chosen model and hardware with representative buffering and concurrent streams
   • Treat one published single-workload number as a guarantee for every deployment
   • Measure only the time needed to download the checkpoint
   • Ignore audio-chunk settings because they cannot affect latency