API Benchmark

The benchmark below reflects the performance metrics from approximately a 3-minute dialogue comprising diverse questions about blockchain and AI. For each test, we established a web socket connection to interact with various model APIs based on the web-app that has already been developed. Note: only the models with streaming features have been benchmarked (excluding OpenAI’s Whisper).

image

Deepgram Nova 2

Average Duration (per transcript): 2977.38 ms

Average Latency Per Word: 540.81 ms

Total Duration: 125050 ms Longest Transcript Duration: 4930 ms (Words in Longest Transcript: 12, Duration for Longest Word Count: 4730 ms) Shortest Transcript Duration: 950 ms (Words in Shortest Transcript: 1, Duration for Shortest Word Count: 1310 ms) Total Number of Transcripts: 42 Total Number of Words: 260 Average Word Count: 6.19

Deepgram ConversationalAI

Average Duration (per transcript): 3302.82 ms

Average Latency Per Word: 515.72 ms

Total Duration: 128810 ms Longest Transcript Duration: 4970 ms (Words in Longest Transcript: 12, Duration for Longest Word Count: 4900 ms) Shortest Transcript Duration: 700 ms (Words in Shortest Transcript: 1, Duration for Shortest Word Count: 700 ms) Total Number of Transcripts: 39 Total Number of Words: 272 Average Word Count: 6.97

Microsoft Azure

Average Duration (per transcript): 5146.05 ms

Average Latency Per Word: 796.34 ms

Total Duration: 210978 ms Longest Transcript Duration: 10085 ms (Words in Longest Transcript: 13, Duration for Longest Word Count: 10250 ms) Shortest Transcript Duration: 1900 ms (Words in Shortest Transcript: 1, Duration for Shortest Word Count: 1900 ms) Total Number of Transcripts: 41 Total Number of Words: 265 Average Word Count: 6.46

Speechmatics

Average Duration (per transcript): 6547.20 ms

Average Latency Per Word: 1041.55 ms

Total Duration: 275110 ms Longest Transcript Duration: 10867 ms (Words in Longest Transcript: 12, Duration for Longest Word Count: 10435 ms) Shortest Transcript Duration: 2090 ms (Words in Shortest Transcript: 1, Duration for Shortest Word Count: 2887 ms) Total Number of Transcripts: 42 Total Number of Words: 264 Average Word Count: 6.29

Non-streaming Speech to Text

Note that non-streaming speech to text is different as the audio file is uploaded and processed. Rather, than users speaking into the media and streams being generated (which inevitably causes streaming to be slower).

OpenAI Whisper

Total Latency: 6129.27 ms

Result:

"What are the primary challenges in integrating AI with blockchain technology? How can blockchain technology ensure the authenticity of the data used by AI algorithms? In what ways can AI improve the functionality and reliability of smart contracts on a blockchain? How might we leverage blockchain to create decentralized AI applications? Can integrating AI with blockchain help enhance user privacy, and if so, how? How can AI be utilized to optimize blockchain network management and operations? Does the integration of AI within blockchain networks introduce any specific security vulnerabilities? Can AI help blockchain scale more efficiently, particularly regarding transaction speeds and data storage? How can AI and blockchain work together to ensure regulatory compliance in industries like finance or healthcare? How is AI used in predictive analysis for cryptocurrency markets? What are the potential benefits and risks of tokenizing AI models on the blockchain? How can blockchain's transparency help address ethical concerns and bias in AI algorithms? Can blockchain technology make the training of AI models more secure or reliable? How could we use blockchain and AI to create decentralized marketplaces for data and algorithms? How can blockchain be used to create immutable audit trails for AI decision making? How can AI algorithms enhance fraud detections and blockchain transactions? What role does AI play in managing and reducing the energy consumption of blockchain networks? What are the implications of AI-generated art or content being sold as NFTs on the blockchain? Can AI be used to make decentralized autonomous organizations more efficient or autonomous? What are some cutting-edge developments at the intersection of AI and blockchain that could transform industries?”