Yakha isixazululo se-audio esingenama-audio nge-Amazon Bedrock kanye nokuhleba

0 0 6 minutes read

Yakha isixazululo se-audio esingenama-audio nge-Amazon Bedrock kanye nokuhleba

Ukuqoshwa kwemihlangano yebhizinisi, izingxoxo, kanye nokusebenzisana kwamakhasimende sekubalulekile ukulondolozwa imininingwane ebalulekile. Kodwa-ke, ukuphrinta nokufingqa lokhu kuqoshwa ngesandla kuvame ukuchitha isikhathi nokusebenza kanzima. Ngentuthuko ekwakhekeni kwe-AI ne-automatic speech submeals (ASR), izixazululo ezenzakalelayo ziye zaqhamuka ukwenza le nqubo isheshe futhi isebenze kakhudlwana.

Ukuvikela imininingwane ekhonjwayo uqobo (i-PII) isici esibalulekile sokuphepha kwedatha, okuqhutshwa yimibili imithwalo yemfanelo yokuziphatha kanye nezidingo zomthetho. Kulokhu okuthunyelwe, sikhombisa ukuthi singalisebenzisa kanjani imodeli evulekile ye-Ai Whiyer (FM) I-V3 Turbo, etholakala e-Amazon Bedrock MarketProce, enikeza ukufinyelela kumamodeli angaphezu kuka-140 ngomnikelo wokuzinikele, ukukhiqiza eduze kwe-Real-Time Praple. Lezi zibhaliwe zibe sezicutshungulwa yi-Amazon Bedrock yokufingqa kanye nokunciphiswa kwemininingwane ebucayi.

I-Amazon Bedrock iyinsizakalo ephethwe ngokuphelele enikeza ukukhetha kwama-FM asebenze kakhulu kusuka ezinkampanini ze-AI ezihamba phambili njenge-AI21 labs, i-anthron, i-Amazon, i-Amazon, i-AI ebanzi, kanye ne-ai ebanzi yamakhono ezokuphepha, ubumfihlo, kanye ne-AI. Ngokwengeziwe, ungasebenzisa i-Amazon Bedrock Guardrails ukuze uthole imininingwane ebucayi ngokuzenzakalelayo, kufaka phakathi i-PII, kusuka ezifingqiwe zokubhaliwe ukusekela ukuhambisana nokuvikelwa kwedatha.

Kulokhu okuthunyelwe, sihamba ngokwakha okuphelele kokuphela okuhlanganisa i-archend-esekwe yi-bedrock esekwe e-Amazon, ama-AWS Lambda, kanye nemisebenzi yezinyathelo ze-AWS ukuhlela ukuhlanganiswa komsebenzi, ukwenza lula ukuhlanganiswa komgwaqo kanye nokusebenza.

Ukubuka konke

Isixazululo siqokomisa amandla okuhlanganisa ubuchwepheshe obungenasici nge-ai generative ai ukuze ushintshe futhi ulinganise ukuguqulwa kokusebenza kokuqukethwe. Uhambo lomsebenzisi luqala ngokulayisha okurekhodiwe ngohlelo lokusebenza lwe-Accempron Flovend, lusingathwe ku-Amazon Cloudfront futhi luxhaswe yi-Amazon Simple Storage Service (Amazon S3) ne-Amazon API Gateway. Lapho ifayela lilayishwa, lidala i-Step Machine State Machine ohlela izinyathelo zokucubungula ezisemqoka, usebenzisa amamodeli we-AI nemisebenzi yeLambda yokugeleza kwedatha engenamthungo. Umdwebo olandelayo ukhombisa ukwakhiwa kwesixazululo.

Ukuchitheka komsebenzi kuqukethe lezi zinyathelo ezilandelayo:

Isicelo se-Act Teact sisingathwe kubhakede le-S3 futhi sasebenza kubasebenzisi ngefufloffront for Fast, Ukufinyelela Komhlaba. I-API Gateway Hambela ukusebenzisana phakathi kwe-Frontlend kanye ne-Backund Services.
Abasebenzisi balayisha amafayela alalelwayo noma wevidiyo ngokuqondile kusuka kuhlelo lokusebenza. Lokhu okuqoshwa kugcinwa kubhakede elibekiwe le-S3 ukucubungula.
Umthetho we-IzinkundlaBridbridge Rule wawuthola umcimbi wokulayisha we-S3 futhi ukhiphe izinyathelo zemisebenzi yomshini wombuso, eqala ipayipi elisebenza nge-AI-Powered Powered.
Umshini wombuso wenza okubhaliwe okulalelwayo, ukufingqa, kanye nokubhekisisa ngokulandelana kwamamodeli ama-Amazon Bedrock amaningi ngokulandelana. Isebenzisa ukuhleba okubhaliwe, i-Claude yokufingqa, kanye nama-Guardrails ukuze uthole imininingwane ebucayi.
Isifinyezo esisenziwe kabusha sibuyiselwa kuhlelo lokusebenza lwe-frontlend futhi luboniswe kumsebenzisi.

Umdwebo olandelayo ukhombisa ukusebenza komshini wombuso.

Yakha isixazululo se-audio esingenama-audio nge-Amazon Bedrock kanye nokuhleba

Isinyathelo sisebenza ngomshini wombuso ohlobana nochungechunge lwemisebenzi yokubhala, ukufingqe, kanye nemininingwane ebucayi ekuqoqweni okulayishiwe okulayishiwe:

Umsebenzi we-lambda ubangelwa ukuqoqa imininingwane yokufaka (ngokwesibonelo, indlela yento ye-Amazon S3, i-metadata) bese ulungiselela ukulayisha okubhaliwe.
I-Payload ithunyelwa kumodeli enkulu ye-OpenAi Turbo Turbo nge-Imazon Bedrock Market ukukhiqiza okubhalwe ngakusasa kwesikhathi sokuqoshwa.
Umbhalo ovundwayo udluliselwe e-Anthropic's Claude Sonnet 3.5 nge-Amazon Bedrock, ekhiqiza isifinyezo esifushane nesibumbano sengxoxo noma okuqukethwe.
Umsebenzi wesibili we-lambda uqinisekisa futhi udlulisela phambili isifinyezo esiteji sokunciphisa.
Isifinyezo sicutshungulwa nge-Amazon Bedrock Guardrails, ehlehla ngokuzenzakalelayo i-PII neminye imininingwane ebucayi.
Isifinyezo esisenziwe kabusha sigcinwa noma sibuyiselwe kuhlelo lokusebenza lwe-Frontlend ngokusebenzisa i-API, lapho kuboniswa khona umsebenzisi.

Izimfuneko

Ngaphambi kokuthi uqale, qiniseka ukuthi unezimfuneko ezilandelayo endaweni:

Dala i-Guardrail e-Amazon Bedrock Console

Ngemiyalo yokwakha ama-Gudgerails e-Amazon Bedrock, bheka ukudala i-Guardrail. Ngemininingwane ekutholeni nasekunciphiseni i-PII, bheka Susa i-PII kusuka ezingxoxweni ngokusebenzisa izihlungi zolwazi olubucayi. Lungiselela i-GuardRail yakho ngezilungiselelo ezilandelayo ezilandelayo:

Nika amandla ukutholwa kwe-PII nokuphatha
Setha isenzo se-PII ukuze uphinde wenze kabusha
Faka izinhlobo ze-PII ezifanele, ezinjengokuthi:
- Amagama namamazisi
- Izinombolo zocingo
- Amakheli E-imeyili
- Amakheli abonakalayo
- Imininingwane yezezimali
- Eminye imininingwane yomuntu siqu ebucayi

Ngemuva kokuthumela i-Guarderail, qaphela igama lesisetshenziswa se-Amazon (ARN), futhi uzobe usebenzisa lokhu lapho ubeka imodeli.

Faka imodeli yokuhleba

Qedela lezi zinyathelo ezilandelayo zokuthumela imodeli enkulu ye-V3 Turbo ebalulekile:

E-Amazon Bedrock Console, khetha Ikhathalogi Model ngaphansi kwa- Amamodeli weSisekelo kufasitelana lokuzulazula.
Sesha bese ukhetha Whisper enkulu v3 turbo.
Kumenyu yezinketho (amachashazi amathathu), khetha Khankasa.

I-Amazon Bedrock Consock Ukubonisa ikhathalogi egcinwe ehlungiwe ngemodeli enkulu ye-V3 Turbo Yenkulumo Yokuqashelwa kanye Nenketho Yokuhambisa

Shintsha igama le-Endpoint, inombolo yezimo, kanye nohlobo oluthile ukuze ivumelane icala lakho lokusebenzisa. Kulokhu okuthunyelwe, sisebenzisa izilungiselelo ezizenzakalelayo.
Shintsha le Izilungiselelo ezithuthukisiwe isigaba ukuze sivumelane icala lakho lokusebenzisa. Kulokhu okuthunyelwe, sisebenzisa izilungiselelo ezizenzakalelayo.
Qoka Khankasa.

Lokhu kudala ubunikazi obusha be-aw.

Ungakhetha Ukuhanjiswa Kwezimakethe kufasitelana navigation, naku Ukuphathwa okuphethwe isigaba, ungabona isimo se-Endpoint njengoba Ukudala. Linda ukuphela kokuqeda ukuthunyelwa kanye nesimo sokushintsha ukuze Enkonzwenibese ukopisha igama le-Endpoint, futhi uzobe usebenzisa lokhu lapho uthumela

I-Amazon Bedrock Console: "Isebenza kanjani" Sibutsetelo, ukuphatha ukuthunyelwa kwetafula nge-Whisper Model Endpoint in Service

Faka ingqalasizinda yesisombululo

E-GitHub Repo, landela imiyalo kufayela le-Readme ukuze uhlanganise okugciniwe, bese ufaka i-frondlend bese ubeka emuva ingqalasizinda.

Sisebenzisa i-AWS Cloud Development Kit (AWS CDK) ukuchaza nokuthumela ingqalasizinda. Ikhodi ye-AWS CDK isebenzisa izinsiza ezilandelayo:

Isicelo se-Reaction Frontlend
Buyisela emuva ingqalasizinda
Amabhakede e-S3 okugcina ukulayishwa kanye nemiphumela esetshenzisiwe
Isinyathelo semisebenzi umshini wombuso nemisebenzi ye-lambda yokucubungula okulalelwayo kanye nokuhlelwa kwe-PII
I-API Gateway Endpoints yokuphatha izicelo
Izindima nezinqubomgomo zokufinyelela okuphephile
Ukusatshalaliswa kwamafu lokusingathwa kwe-frontlend

Ukuqalisa Ukusetshenziswa okujulile

I-Backend yakhiwa ngokulandelana kwemisebenzi ye-lambda, ngakunye ukuphatha isigaba esithile sepayipi elinomsindo:

Faka isiphanzi – Ithola amafayela alalelwayo futhi uwagcina e-Amazon S3
Ukubhalwa ngokushwabana – Ukuguqula inkulumo ibe umbhalo usebenzisa imodeli yokuhleba
Ukutholwa kwesipika – Ukwahlukanisa kanye namalebula izikhulumi ngazinye ngaphakathi komsindo
Ukufingqa kusetshenziswa i-Amazon Bedrock – Ukukhishwa bese kufingqa amaphuzu asemqoka kusuka ku-transcript
Ukulahlekelwa kwePii – Isebenzisa i-Amazon Bedrock Guardrails ukususa imininingwane ebucayi yokuhambisana nobumfihlo

Ake sihlole ezinye zezakhi ezibalulekile:

Umsebenzi wokubhala we-Lambda usebenzisa imodeli yokuhlenga ukuguqula amafayela alalelwayo abe umbhalo:

def transcribe_with_whisper(audio_chunk, endpoint_name):
    # Convert audio to hex string format
    hex_audio = audio_chunk.hex()
    
    # Create payload for Whisper model
    payload = {
        "audio_input": hex_audio,
        "language": "english",
        "task": "transcribe",
        "top_p": 0.9
    }
    
    # Invoke the SageMaker endpoint running Whisper
    response = sagemaker_runtime.invoke_endpoint(
        EndpointName=endpoint_name,
        ContentType="application/json",
        Body=json.dumps(payload)
    )
    
    # Parse the transcription response
    response_body = json.loads(response['Body'].read().decode('utf-8'))
    transcription_text = response_body['text']
    
    return transcription_text

Sisebenzisa i-Amazon Bedrock ukukhiqiza izifingqo ezimfushane kusuka kokubhaliwe:

def generate_summary(transcription):
    # Format the prompt with the transcription
    prompt = f"{transcription}nnGive me the summary, speakers, key discussions, and action items with owners"
    
    # Call Bedrock for summarization
    response = bedrock_runtime.invoke_model(
        modelId="anthropic.claude-3-5-sonnet-20240620-v1:0",
        body=json.dumps({
            "prompt": prompt,
            "max_tokens_to_sample": 4096,
            "temperature": 0.7,
            "top_p": 0.9,
        })
    )
    
    # Extract and return the summary
    result = json.loads(response.get('body').read())
    return result.get('completion')

Ingxenye ebucayi yesisombululo sethu ukuncipha okuzenzakalelayo kwe-PII. Sisebenzise lokhu ngokusebenzisa i-Amazon Bedrock Guardrails ukusekela ukuhambisana nemithetho yobumfihlo:

def apply_guardrail(bedrock_runtime, content, guardrail_id):
# Format content according to API requirements
formatted_content = [{"text": {"text": content}}]

# Call the guardrail API
response = bedrock_runtime.apply_guardrail(
guardrailIdentifier=guardrail_id,
guardrailVersion="DRAFT",
source="OUTPUT",  # Using OUTPUT parameter for proper flow
content=formatted_content
)

# Extract redacted text from response
if 'action' in response and response['action'] == 'GUARDRAIL_INTERVENED':
if len(response['outputs']) > 0:
output = response['outputs'][0]
if 'text' in output and isinstance(output['text'], str):
return output['text']

# Return original content if redaction fails
return content

Lapho i-PII itholwa, ithathelwe indawo yezinkomba zohlobo (ngokwesibonelo, {ifoni} noma {imeyili}), ukuqiniseka ukuthi izifingqo zihlala zinolwazi ngenkathi zivikela idatha ebucayi.

Ukuphatha ipayipi elinzima lokucubungula, sisebenzisa imisebenzi yesinyathelo ukuhlela imisebenzi ye-lambda:

{
"Comment": "Audio Summarization Workflow",
"StartAt": "TranscribeAudio",
"States": {
"TranscribeAudio": {
"Type": "Task",
"Resource": "arn:aws:states:::lambda:invoke",
"Parameters": {
"FunctionName": "WhisperTranscriptionFunction",
"Payload": {
"bucket": "$.bucket",
"key": "$.key"
}
},
"Next": "IdentifySpeakers"
},
"IdentifySpeakers": {
"Type": "Task",
"Resource": "arn:aws:states:::lambda:invoke",
"Parameters": {
"FunctionName": "SpeakerIdentificationFunction",
"Payload": {
"Transcription.$": "$.Payload"
}
},
"Next": "GenerateSummary"
},
"GenerateSummary": {
"Type": "Task",
"Resource": "arn:aws:states:::lambda:invoke",
"Parameters": {
"FunctionName": "BedrockSummaryFunction",
"Payload": {
"SpeakerIdentification.$": "$.Payload"
}
},
"End": true
}
}
}

Lokhu kuhamba komsebenzi kuqinisekisa ukuthi isinyathelo ngasinye siqeda ngempumelelo ngaphambi kokuqhubeka nesheke ngokulandelayo, nge-othomathikhi ngephutha lokuphatha bese uzama kabusha logic owakhiwe ngaphakathi.

Hlola ikhambi

Ngemuva kokuthi usuqede ukuthunyelwa ngempumelelo, ungasebenzisa i-URL ye-Cloudfront ukuvivinya ukusebenza kwesisombululo.

Ukulayisha okulalelwayo / ividiyo nokuxhumeka kwesifinyezo ngokulayishwa kwefayela okuqediwe kokuhlaziywa kokuqoshwa komhlangano weqembu

Ukucatshangelwa Kwezokuphepha

Ezokuphepha ziyisici esibucayi salesi sixazululo, futhi sisebenzise izindlela eziningana ezingcono kakhulu zokusekela ukuvikela idatha nokuhambisana:

Ukuhlelwa kwedatha ebucayi – Yenza kabusha i-PII ngokuzenzakalelayo ukuvikela ubumfihlo bomsebenzisi.
Izimvume ze-IAM ezihlanzekile – Sebenzisa umgomo wokuthola ilungelo okungenani izinsizakalo ze-AWS nezinsizakusebenza.
Izilawuli Zokufinyelela ze-Amazon S3 – Sebenzisa izinqubomgomo eziqinile zebhakede ukukhawulela ukufinyelela kubasebenzisi abagunyaziwe nezindima.
E-API Security – Ukuvikeleka kwe-API Endpoints usebenzisa i-Amazon Cognito ngokuqinisekiswa komsebenzisi (ngokuzithandela kepha kunconyiwe).
Ukuvikelwa kwamafu – Eforce HTTSPS bese usebenzisa amaphrothokhokeli anamuhla we-TLS ukwenza lula ukulethwa okuqukethwe okuphephile.
I-Amazon Bedrock Idatha Yezokuphepha – I-Amazon Bedrock (kufaka phakathi i-Amazon Bedrock MarketPlace) Ivikela idatha yamakhasimende futhi ayithumeli idatha kubahlinzeki noma isitimela usebenzisa idatha yamakhasimende. Lokhu kuqinisekisa ukuthi imininingwane yakho yokuphathelene ihlala iphephile lapho usebenzisa amakhono e-AI.

Hlanza

Ukuvimbela amacala angadingekile, qiniseka ukuthi ususa izinsizakusebenza ezihlinzekelwe lesi sixazululo uma usuqedile:

Susa i-Amazon Bedrock Guardral:
1. E-Amazon Bedrock Console, kwimenyu yokuhambisa, khetha Ukuqapha.
2. Khetha i-Guardrail yakho, bese ukhetha Cisha.
Susa imodeli enkulu ye-v3 turbo esetshenzisiwe nge-Amazon Bedrock MarketPlace:
1. E-Amazon Bedrock Console, khetha Ukuhanjiswa Kwezimakethe kufasitelana lokuzulazula.
2. Ku Ukuphathwa okuphethwe Ingxenye, khetha isiphetho esithunyelwe bese ukhetha Cisha.
Susa isitaki se-AWS CDK ngokusebenza komyalo cdk destroyesusa ingqalasizinda yezempi.

Ukugcina

Lesi sixazululo esifingqiwe se-Spreadless sibonisa izinzuzo zokuhlanganisa izinsizakalo ze-AWS ukudala uhlelo oluyinkimbinkimbi, oluphephile, futhi olunakwa. Ngokusebenzisa i-Amazon Bedrock yamakhono e-AI, iLambda yokucubungula kusengaphambili, kanye ne-Cloudfront yokulethwa kokuqukethwe, sakha isixazululo esingaphatha amavolumu amakhulu okuqukethwe okulalelwayo kahle ngenkathi kukusiza ukuvumelanisa nemikhuba engcono yokuphepha.

Isici sokuhlelwa kabusha kwe-othomathikhi ye-PII sisekela ukuhambisana nemigomo yobumfihlo, okwenza lesi sixazululo silungele kahle izimboni ezilawulwayo ezifana nezokunakekelwa kwempilo, ezezimali, kanye nezinsizakalo zezomthetho lapho ukuphepha kwedatha lapho kukhulu khona ukuphepha kwedatha. Ukuze uqalise, faka lolu daba lwezakhiwo ngaphakathi kwemvelo yakho ye-AWS ukusheshisa ukuhamba komsindo kwakho kokulalelwayo.

Mayelana nababhali

Kaiyin Hu Ingabe ukwakhiwa kwezixazululo okuphezulu kwama-akhawunti amasu e-Amazon Web Services, eneminyaka yolwazi kuwo wonke amabhizinisi, ama-Startups, kanye nezinsizakalo eziqeqeshiwe. Njengamanje, usiza amakhasimende ekwakhelwe izixazululo zamafu futhi adonsele ukwamukelwa kweGenai efu. Phambilini, uKaiyin wasebenza esizindeni sasekhaya esihlakaniphile, ukusiza amakhasimende ekuhlanganiseni ubuchwepheshe bezwi ne-IOT.

Sid Vantair Ingabe ukwakhiwa kwezixazululo ngama-AW ana aze amboze amasu. Uyachuma ekuxazululeni izingqinamba zobuchwepheshe eziyinkimbinkimbi ukunqoba izithiyo zamakhasimende. Ngaphandle komsebenzi, uyakwazisa ukuchitha isikhathi nomndeni wakhe nokukhuthaza ubukeke ezinganeni zakhe.